Search Results

Search found 5954 results on 239 pages for 'cpu cores'.

Page 77/239 | < Previous Page | 73 74 75 76 77 78 79 80 81 82 83 84  | Next Page >

  • x86 linux - how to create custom malloc with address hint

    - by nandu
    Hi, I want to create a custom malloc which allocates memory blocks within a given address range. I am writing a pthreads application in which threads are bound to unique cores on a many-core machine. The memory controllers are statically mapped, so that certain range of addresses on main memory are electrically closer to a core. I want to minimize the latency of communication between cores and main memory by allocating thread memory on these "closer" regions. Any ideas would be most appreciated. Thank you! Nandu

    Read the article

  • Simulators for thread scheduling on multicore

    - by shijie xu
    I am seeking a simulator for thread scheduling at multi-core architecture, that is mapping threads to the cores at runtime. During runtime, simulator collects overall cache and IPC statistics. I checked below simulators, but seems there are not sufficient for me: Simplescalar: A simulator only for single core. SESC: multiprocessor simulator with detailed power, thermal, and performance models, QSim: provides instruction-level control of the emulated environment and detailed information about the executing instruction stream. It seems both SESC and QSim supports instructions scheduling instead of thread scheduling on the cores? Anyone can help provide some clues or share experience for this part?

    Read the article

  • SElinux process killed while trying to set boolean

    - by Antonio
    I've got a strange problem. I can not allow apache to connect to database at my CentOC 6.4 box: [root@centos6 ~]# setsebool -P httpd_can_network_connect on Killed [root@centos6 ~]# sestatus -b | grep httpd_can_network_connect httpd_can_network_connect off httpd_can_network_connect_cobbler off httpd_can_network_connect_db off I watched log file, but there was no log messages: tail -f /var/log/audit/audit.log UPDATE: There are some information in /var/log/messages: Nov 9 19:07:16 vs302 kernel: setsebool invoked oom-killer: gfp_mask=0x280da, order=0, oom_adj=0, oom_score_adj=0 Nov 9 19:07:16 vs302 kernel: setsebool cpuset=/ mems_allowed=0 Nov 9 19:07:16 vs302 kernel: Pid: 1660, comm: setsebool Not tainted 2.6.32-358.23.2.el6.x86_64 #1 Nov 9 19:07:16 vs302 kernel: Call Trace: Nov 9 19:07:16 vs302 kernel: [<ffffffff810cb641>] ? cpuset_print_task_mems_allowed+0x91/0xb0 Nov 9 19:07:16 vs302 kernel: [<ffffffff8111ce40>] ? dump_header+0x90/0x1b0 Nov 9 19:07:16 vs302 kernel: [<ffffffff8111d2c2>] ? oom_kill_process+0x82/0x2a0 Nov 9 19:07:16 vs302 kernel: [<ffffffff8111d201>] ? select_bad_process+0xe1/0x120 Nov 9 19:07:16 vs302 kernel: [<ffffffff8111d700>] ? out_of_memory+0x220/0x3c0 Nov 9 19:07:16 vs302 kernel: [<ffffffff8112c3dc>] ? __alloc_pages_nodemask+0x8ac/0x8d0 Nov 9 19:07:16 vs302 kernel: [<ffffffff81160d6a>] ? alloc_pages_vma+0x9a/0x150 Nov 9 19:07:16 vs302 kernel: [<ffffffff81143f0b>] ? handle_pte_fault+0x76b/0xb50 Nov 9 19:07:16 vs302 kernel: [<ffffffff81228664>] ? task_has_capability+0xb4/0x110 Nov 9 19:07:16 vs302 kernel: [<ffffffff81004a49>] ? __raw_callee_save_xen_pmd_val+0x11/0x1e Nov 9 19:07:16 vs302 kernel: [<ffffffff8114452a>] ? handle_mm_fault+0x23a/0x310 Nov 9 19:07:16 vs302 kernel: [<ffffffff811485b6>] ? vma_adjust+0x556/0x5e0 Nov 9 19:07:16 vs302 kernel: [<ffffffff810474e9>] ? __do_page_fault+0x139/0x480 Nov 9 19:07:16 vs302 kernel: [<ffffffff81148b8a>] ? vma_merge+0x29a/0x3e0 Nov 9 19:07:16 vs302 kernel: [<ffffffff81149fdc>] ? do_brk+0x26c/0x350 Nov 9 19:07:16 vs302 kernel: [<ffffffff8100ba1d>] ? retint_restore_args+0x5/0x6 Nov 9 19:07:16 vs302 kernel: [<ffffffff81513bfe>] ? do_page_fault+0x3e/0xa0 Nov 9 19:07:16 vs302 kernel: [<ffffffff81510fb5>] ? page_fault+0x25/0x30 Nov 9 19:07:16 vs302 kernel: Mem-Info: Nov 9 19:07:16 vs302 kernel: Node 0 DMA per-cpu: Nov 9 19:07:16 vs302 kernel: CPU 0: hi: 0, btch: 1 usd: 0 Nov 9 19:07:16 vs302 kernel: Node 0 DMA32 per-cpu: Nov 9 19:07:16 vs302 kernel: CPU 0: hi: 186, btch: 31 usd: 30 Nov 9 19:07:16 vs302 kernel: active_anon:132249 inactive_anon:46 isolated_anon:0 Nov 9 19:07:16 vs302 kernel: active_file:56 inactive_file:59 isolated_file:0 Nov 9 19:07:16 vs302 kernel: unevictable:0 dirty:2 writeback:0 unstable:0 Nov 9 19:07:16 vs302 kernel: free:1369 slab_reclaimable:1774 slab_unreclaimable:11588 Nov 9 19:07:16 vs302 kernel: mapped:54 shmem:48 pagetables:1211 bounce:0 Nov 9 19:07:16 vs302 kernel: Node 0 DMA free:2440kB min:72kB low:88kB high:108kB active_anon:12156kB inactive_anon:0kB active_file:0kB inactive_file:0kB unevictable:0kB isolated(anon):0kB isolated(file):0kB present:14648kB mlocked:0kB dirty:0kB writeback:0kB mapped:0kB shmem:0kB slab_reclaimable:24kB slab_unreclaimable:8kB kernel_stack:0kB pagetables:16kB unstable:0kB bounce:0kB writeback_tmp:0kB pages_scanned:0 all_unreclaimable? yes Nov 9 19:07:16 vs302 kernel: lowmem_reserve[]: 0 590 590 590 Nov 9 19:07:16 vs302 kernel: Node 0 DMA32 free:3036kB min:3072kB low:3840kB high:4608kB active_anon:516840kB inactive_anon:184kB active_file:224kB inactive_file:236kB unevictable:0kB isolated(anon):0kB isolated(file):0kB present:604988kB mlocked:0kB dirty:8kB writeback:0kB mapped:216kB shmem:192kB slab_reclaimable:7072kB slab_unreclaimable:46344kB kernel_stack:880kB pagetables:4828kB unstable:0kB bounce:0kB writeback_tmp:0kB pages_scanned:128 all_unreclaimable? no Nov 9 19:07:16 vs302 kernel: lowmem_reserve[]: 0 0 0 0 Nov 9 19:07:16 vs302 kernel: Node 0 DMA: 0*4kB 1*8kB 0*16kB 0*32kB 0*64kB 1*128kB 1*256kB 0*512kB 0*1024kB 1*2048kB 0*4096kB = 2440kB Nov 9 19:07:16 vs302 kernel: Node 0 DMA32: 129*4kB 67*8kB 30*16kB 19*32kB 6*64kB 2*128kB 1*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB = 3036kB Nov 9 19:07:16 vs302 kernel: 182 total pagecache pages Nov 9 19:07:16 vs302 kernel: 0 pages in swap cache Nov 9 19:07:16 vs302 kernel: Swap cache stats: add 0, delete 0, find 0/0 Nov 9 19:07:16 vs302 kernel: Free swap = 0kB Nov 9 19:07:16 vs302 kernel: Total swap = 0kB Nov 9 19:07:16 vs302 kernel: 157439 pages RAM Nov 9 19:07:16 vs302 kernel: 6271 pages reserved Nov 9 19:07:16 vs302 kernel: 2686 pages shared Nov 9 19:07:16 vs302 kernel: 146395 pages non-shared Nov 9 19:07:16 vs302 kernel: [ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name Nov 9 19:07:16 vs302 kernel: [ 271] 0 271 2798 231 0 -17 -1000 udevd Nov 9 19:07:16 vs302 kernel: [ 476] 0 476 2797 230 0 -17 -1000 udevd Nov 9 19:07:16 vs302 kernel: [ 718] 0 718 2279 122 0 0 0 dhclient Nov 9 19:07:16 vs302 kernel: [ 762] 0 762 6909 58 0 -17 -1000 auditd Nov 9 19:07:16 vs302 kernel: [ 787] 0 787 62270 147 0 0 0 rsyslogd Nov 9 19:07:16 vs302 kernel: [ 801] 25 801 40326 2655 0 0 0 named Nov 9 19:07:16 vs302 kernel: [ 850] 0 850 16563 172 0 -17 -1000 sshd Nov 9 19:07:16 vs302 kernel: [ 875] 0 875 23451 240 0 0 0 sshd Nov 9 19:07:16 vs302 kernel: [ 966] 498 966 4780 44 0 0 0 wrapper Nov 9 19:07:16 vs302 kernel: [ 968] 498 968 497404 40812 0 0 0 java Nov 9 19:07:16 vs302 kernel: [ 1057] 0 1057 20216 225 0 0 0 master Nov 9 19:07:16 vs302 kernel: [ 1064] 89 1064 20278 209 0 0 0 qmgr Nov 9 19:07:16 vs302 kernel: [ 1071] 0 1071 27075 121 0 0 0 bash Nov 9 19:07:16 vs302 kernel: [ 1111] 0 1111 24880 350 0 0 0 httpd Nov 9 19:07:16 vs302 kernel: [ 1117] 48 1117 24913 351 0 0 0 httpd Nov 9 19:07:16 vs302 kernel: [ 1118] 48 1118 24880 337 0 0 0 httpd Nov 9 19:07:16 vs302 kernel: [ 1119] 48 1119 24880 337 0 0 0 httpd Nov 9 19:07:16 vs302 kernel: [ 1120] 48 1120 24880 337 0 0 0 httpd Nov 9 19:07:16 vs302 kernel: [ 1121] 48 1121 24880 337 0 0 0 httpd Nov 9 19:07:16 vs302 kernel: [ 1122] 48 1122 24880 337 0 0 0 httpd Nov 9 19:07:16 vs302 kernel: [ 1124] 48 1124 24880 337 0 0 0 httpd Nov 9 19:07:16 vs302 kernel: [ 1125] 48 1125 24880 337 0 0 0 httpd Nov 9 19:07:16 vs302 kernel: [ 1129] 0 1129 29313 151 0 0 0 crond Nov 9 19:07:16 vs302 kernel: [ 1143] 0 1143 1018 22 0 0 0 agetty Nov 9 19:07:16 vs302 kernel: [ 1146] 0 1146 1015 22 0 0 0 mingetty Nov 9 19:07:16 vs302 kernel: [ 1514] 0 1514 23451 237 0 0 0 sshd Nov 9 19:07:16 vs302 kernel: [ 1517] 0 1517 27075 113 0 0 0 bash Nov 9 19:07:16 vs302 kernel: [ 1641] 89 1641 20236 218 0 0 0 pickup Nov 9 19:07:16 vs302 kernel: [ 1659] 0 1659 25234 39 0 0 0 tail Nov 9 19:07:16 vs302 kernel: [ 1660] 0 1660 89903 85712 0 0 0 setsebool Nov 9 19:07:16 vs302 kernel: Out of memory: Kill process 1660 (setsebool) score 568 or sacrifice child Nov 9 19:07:16 vs302 kernel: Killed process 1660, UID 0, (setsebool) total-vm:359612kB, anon-rss:342708kB, file-rss:140kB

    Read the article

  • my server suddenly crashes every 2 days or so. Programmer has no idea, please help find the cause, here is the top

    - by Alex
    Every couple of days my server suddenly crashes and I must request hardware reset at data center to get it back running. Today I came back to my shell and saw the server was dead and "top" was running on it, and see below for the "top" right before the crash. I opened /var/log/messages and scrolled to the reboot time and see nothing, no errors prior to the hard reboot. (I checked in /etc/syslog.conf and I see "*.info;mail.none;authpriv.none;cron.none /var/log/messages" , isn't this good enough to log all problems?) Usually when I look at the top, the swap is never used up like this! I also don't know why mysqld is at 323% cpu (server only runs drupal and its never slow or overloaded). Solver is my application. I don't know whats that 'sh' doing and 'dovecot' doing. Its driving me crazy over the last month, please help me solve this mystery and stop my downtimes. top - 01:10:06 up 6 days, 5 min, 3 users, load average: 34.87, 18.68, 9.03 Tasks: 500 total, 19 running, 481 sleeping, 0 stopped, 0 zombie Cpu(s): 0.0%us, 96.6%sy, 0.0%ni, 1.7%id, 1.8%wa, 0.0%hi, 0.0%si, 0.0%st Mem: 8165600k total, 8139764k used, 25836k free, 428k buffers Swap: 2104496k total, 2104496k used, 0k free, 8236k cached PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 4421 mysql 15 0 571m 105m 976 S 323.5 1.3 9:08.00 mysqld 564 root 20 -5 0 0 0 R 99.5 0.0 2:49.16 kswapd1 25767 apache 19 0 399m 8060 888 D 79.3 0.1 0:06.64 httpd 25781 apache 19 0 398m 5648 492 R 79.0 0.1 0:08.21 httpd 25961 apache 25 0 398m 5700 560 R 76.7 0.1 0:17.81 httpd 25980 apache 25 0 10816 668 520 R 75.0 0.0 0:46.95 sh 563 root 20 -5 0 0 0 D 71.4 0.0 3:12.37 kswapd0 25766 apache 25 0 399m 7256 756 R 69.7 0.1 0:39.83 httpd 25911 apache 25 0 398m 5612 480 R 58.8 0.1 0:17.63 httpd 25782 apache 25 0 440m 38m 648 R 55.2 0.5 0:18.94 httpd 25966 apache 25 0 398m 5640 556 R 55.2 0.1 0:48.84 httpd 4588 root 25 0 74860 596 476 R 53.9 0.0 0:37.90 crond 25939 apache 25 0 2776 172 84 R 48.9 0.0 0:59.46 solver 4575 root 25 0 397m 6004 1144 R 48.6 0.1 1:00.43 httpd 25962 apache 25 0 398m 5628 492 R 47.9 0.1 0:14.58 httpd 25824 apache 25 0 440m 39m 680 D 47.3 0.5 0:57.85 httpd 25968 apache 25 0 398m 5612 528 R 46.6 0.1 0:42.73 httpd 4477 root 25 0 6084 396 280 R 46.3 0.0 0:59.53 dovecot 25982 root 25 0 397m 5108 240 R 45.9 0.1 0:18.01 httpd 25943 apache 25 0 2916 172 8 R 44.0 0.0 0:53.54 solver 30687 apache 25 0 468m 63m 1124 D 42.3 0.8 0:45.02 httpd 25978 apache 25 0 398m 5688 600 R 23.8 0.1 0:40.99 httpd 25983 root 25 0 397m 5272 384 D 14.9 0.1 0:18.99 httpd 935 root 10 -5 0 0 0 D 14.2 0.0 1:54.60 kjournald 25986 root 25 0 397m 5308 420 D 8.9 0.1 0:04.75 httpd 4011 haldaemo 25 0 31568 1476 716 S 5.6 0.0 0:24.36 hald 25956 apache 23 0 398m 5872 644 S 5.6 0.1 0:13.85 httpd 18336 root 18 0 13004 1332 724 R 0.3 0.0 1:46.66 top 1 root 18 0 10372 212 180 S 0.0 0.0 0:05.99 init 2 root RT -5 0 0 0 S 0.0 0.0 0:00.95 migration/0 3 root 34 19 0 0 0 S 0.0 0.0 0:00.01 ksoftirqd/0 4 root RT -5 0 0 0 S 0.0 0.0 0:00.00 watchdog/0 5 root RT -5 0 0 0 S 0.0 0.0 0:00.15 migration/1 6 root 34 19 0 0 0 S 0.0 0.0 0:00 .06 ksoftirqd/1 here is a normal top, when server is working fine: top - 01:50:41 up 21 min, 1 user, load average: 2.98, 2.70, 1.68 Tasks: 271 total, 2 running, 269 sleeping, 0 stopped, 0 zombie Cpu(s): 15.0%us, 1.1%sy, 0.0%ni, 81.4%id, 2.4%wa, 0.1%hi, 0.0%si, 0.0%st Mem: 8165600k total, 2035856k used, 6129744k free, 60840k buffers Swap: 2104496k total, 0k used, 2104496k free, 283744k cached PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 2204 apache 17 0 466m 83m 19m S 25.9 1.0 0:22.16 httpd 11347 apache 15 0 466m 83m 19m S 25.9 1.0 0:26.10 httpd 18204 apache 18 0 481m 97m 19m D 25.2 1.2 0:13.99 httpd 4644 apache 18 0 481m 100m 19m D 24.6 1.3 1:17.12 httpd 4727 apache 17 0 481m 99m 19m S 24.3 1.2 1:10.77 httpd 4777 apache 17 0 482m 102m 21m S 23.6 1.3 1:38.27 httpd 8924 apache 15 0 483m 99m 19m S 22.3 1.3 1:13.41 httpd 9390 apache 18 0 483m 99m 19m S 18.9 1.2 1:05.35 httpd 4728 apache 16 0 481m 101m 19m S 14.3 1.3 1:12.50 httpd 4648 apache 15 0 481m 107m 27m S 12.6 1.4 1:18.62 httpd 24955 apache 15 0 467m 82m 19m S 3.3 1.0 0:21.80 httpd 4722 apache 15 0 503m 118m 19m R 1.7 1.5 1:17.79 httpd 4647 apache 15 0 484m 105m 20m S 1.3 1.3 1:40.73 httpd 4643 apache 16 0 481m 100m 20m S 0.7 1.3 1:11.80 httpd 1561 root 15 0 12900 1264 828 R 0.3 0.0 0:00.54 top 4434 mysql 15 0 496m 55m 4812 S 0.3 0.7 0:06.69 mysqld 4646 apache 15 0 481m 100m 19m S 0.3 1.3 1:25.51 httpd 1 root 18 0 10372 692 580 S 0.0 0.0 0:02.09 init 2 root RT -5 0 0 0 S 0.0 0.0 0:00.03 migration/0 3 root 34 19 0 0 0 S 0.0 0.0 0:00.00 ksoftirqd/0 4 root RT -5 0 0 0 S 0.0 0.0 0:00.00 watchdog/0 5 root RT -5 0 0 0 S 0.0 0.0 0:00.00 migration/1 6 root 34 19 0 0 0 S 0.0 0.0 0:00.00 ksoftirqd/1 7 root RT -5 0 0 0 S 0.0 0.0 0:00.00 watchdog/1 8 root RT -5 0 0 0 S 0.0 0.0 0:00.00 migration/2 9 root 34 19 0 0 0 S 0.0 0.0 0:00.00 ksoftirqd/2 10 root RT -5 0 0 0 S 0.0 0.0 0:00.00 watchdog/2 11 root RT -5 0 0 0 S 0.0 0.0 0:00.00 migration/3 12 root 34 19 0 0 0 S 0.0 0.0 0:00.00 ksoftirqd/3 13 root RT -5 0 0 0 S 0.0 0.0 0:00.00 watchdog/3 14 root RT -5 0 0 0 S 0.0 0.0 0:00.03 migration/4 15 root 34 19 0 0 0 S 0.0 0.0 0:00.00 ksoftirqd/4 16 root RT -5 0 0 0 S 0.0 0.0 0:00.00 watchdog/4 17 root RT -5 0 0 0 S 0.0 0.0 0:00.02 migration/5 18 root 34 19 0 0 0 S 0.0 0.0 0:00.00 ksoftirqd/5 19 root RT -5 0 0 0 S 0.0 0.0 0:00.00 watchdog/5 20 root RT -5 0 0 0 S 0.0 0.0 0:00.01 migration/6 21 root 34 19 0 0 0 S 0.0 0.0 0:00.00 ksoftirqd/6 22 root RT -5 0 0 0 S 0.0 0.0 0:00.00 watchdog/6 23 root RT -5 0 0 0 S 0.0 0.0 0:00.00 migration/7

    Read the article

  • Upgrading Windows 8 boot to VHD to Windows 8.1&ndash;Step by step guide

    - by Liam Westley
    Originally posted on: http://geekswithblogs.net/twickers/archive/2013/10/19/upgrading-windows-8-boot-to-vhd-to-windows-8.1ndashstep-by.aspxBoot to VHD – dual booting Windows 7 and Windows 8 became easy When Windows 8 arrived, quite a few people decided that they would still dual boot their machines, and instead of mucking about with resizing disk partitions to free up space for Windows 8 they decided to use the boot from VHD feature to create a huge hard disc image into which Windows 8 could be installed.  Scott Hanselman wrote this installation guide, while I myself used the installation guide from Ed Bott of ZD net fame. Boot to VHD is a great solution, it achieves a dual boot, can be backed up easily and had virtually no effect on the original Windows 7 partition. As a developer who has dual booted Windows operating systems for years, hacking boot.ini files, the boot to VHD was a much easier solution. Upgrade to Windows 8.1 – ah, you can’t do that on a virtual disk installation (boot to VHD) Last week the final version of Windows 8.1 arrived, and I went into the Windows Store to upgrade.  Luckily I’m on a fast download service, and use an SSD, because once the upgrade was downloaded and prepared Windows informed that This PC can’t run Windows 8.1, and provided the reason, You can’t install Windows on a virtual drive.  You can see an image of the message and discussion that sparked my search for a solution in this Microsoft Technet forum post. I was determined not to have to resize partitions yet again and fiddle with VHD to disk utilities and back again, and in the end I did succeed in upgrading to a Windows 8.1 boot to VHD partition.  It takes quite a bit of effort though … tldr; Simple steps of how you upgrade Boot into Windows 7 – make a copy of your Windows 8 VHD, to become Windows 8.1 Enable Hyper-V in your Windows 8 (the original boot to VHD partition) Create a new virtual machine, attaching the copy of your Windows 8 VHD Start the virtual machine, upgrade it via the Windows Store to Windows 8.1 Shutdown the virtual machine Boot into Windows 7 – use the bcedit tool to create a new Windows 8.1 boot to VHD option (pointing at the copy) Boot into the new Windows 8.1 option Reactivate Windows 8.1 (it will have become deactivated by running under Hyper-V) Remove the original Windows 8 VHD, and in Windows 7 use bcedit to remove it from the boot menu Things you’ll need A system that can run Hyper-V under Windows 8 (Intel i5, i7 class CPU) Enough space to have your original Windows 8 boot to VHD and a copy at the same time An ISO or DVD for Windows 8 to create a bootable Windows 8 partition Step by step guide Boot to your base o/s, the real one, Windows 7. Make a copy of the Windows 8 VHD file that you use to boot Windows 8 (via boot from VHD) – I copied it from a folder on C: called VHD-Win8 to VHD-Win8.1 on my N: drive. Reboot your system into Windows 8, and enable Hyper-V if not already present (this may require reboot) Use the Hyper-V manager , create a new Hyper-V machine, using half your system memory, and use the option to attach an existing VHD on the main IDE controller – this will be the new copy you made in Step 2. Start the virtual machine, use Connect to view it, and you’ll probably discover it cannot boot as there is no boot record If this is the case, go to Hyper-V manager, edit the Settings for the virtual machine to attach an ISO of a Windows 8 DVD to the second IDE controller. Start the virtual machine, use Connect to view it, and it should now attempt a fresh installation of Windows 8.  You should select Advanced Options and choose Repair - this will make VHD bootable When the setup reboots your virtual machine, turn off the virtual machine, and remove the ISO of the Windows 8 DVD from the virtual machine settings. Start virtual machine, use Connect to view it.  You will see the devices to be re-discovered (including your quad CPU becoming single CPU).  Eventually you should see the Windows Login screen. You may notice that your desktop background (Win+D) will have turned black as your Windows installation has become deactivate due to the hardware changes between your real PC and Hyper-V. Fortunately becoming deactivated, does not stop you using the Windows Store, where you can select the update to Windows 8.1. You can now watch the progress joy of the Windows 8 update; downloading, preparing to update, checking compatibility, gathering info, preparing to restart, and finally, confirm restart - remember that you are restarting your virtual machine sitting on the copy of the VHD, not the Windows 8 boot to VHD you are currently using to run Hyper-V (confused yet?) After the reboot you get the real upgrade messages; setting up x%, xx%, (quite slow) After a while, Getting ready Applying PC Settings x%, xx% (really slow) Updating your system (fast) Setting up a few more things x%, (quite slow) Getting ready, again Accept license terms Express settings Confirmed previous password Next, I had to set up a Microsoft account – which is possibly now required, and not optional Using the Microsoft account required a 2 factor authorization, via text message, a 7 digit code for me Finalising settings Blank screen, HI .. We're setting up things for you (similar to original Windows 8 install) 'You can get new apps from the Store', below which is ’Installing your apps’ - I had Windows Media Center which is counts as an app from the Store ‘Taking care of a few things’, below which is ‘Installing your apps’ ‘Taking care of a few things’, below ‘Don't turn off your PC’ ‘Getting your apps ready’, below ‘Don't turn off your PC’ ‘Almost ready’, below ‘Don't turn off your PC’ … finally, we get the Windows 8.1 start menu, and a quick Win+D to check the desktop confirmed all the application icons I expected, pinned items on the taskbar, and one app moaning about a missing drive At this point the upgrade is complete – you can shutdown the virtual machine Reboot from the original Windows 8 and return to Windows 7 to configure booting to the Windows 8.1 copy of the VHD In an administrator command prompt do following use the bcdedit tool (from an MSDN blog about configuring VHD to boot in Windows 7) Type bcedit to list the current boot options, so you can copy the GUID (complete with brackets/braces) for the original Windows 8 boot to VHD Create a new menu option, copy of the Windows 8 option; bcdedit /copy {originalguid} /d "Windows 8.1" Point the new Windows 8.1 option to the copy of the VHD; bcdedit /set {newguid} device vhd=[D:]\Image.vhd Point the new Windows 8.1 option to the copy of the VHD; bcdedit /set {newguid} osdevice vhd=[D:]\Image.vhd Set autodetection of the HAL (may already be set); bcdedit /set {newguid} detecthal on Reboot from Windows 7 and select the new option 'Windows 8.1' on the boot menu, and you’ll have some messages to look at, as your hardware is redetected (as you are back from 1 CPU to 4 CPUs) ‘Getting devices ready, blank then %xx, with occasional blank screen, for the graphics driver, (fast-ish) Getting Ready message (fast) You will have to suffer one final reboots, choose 'Windows 8.1' and you can now login to a lovely Windows 8.1 start screen running on non virtualized hardware via boot to VHD After checking everything is running fine, you can now choose to Activate Windows, which for me was a toll free phone call to the automated system where you type in lots of numbers to be given a whole bunch of new activation codes. Once you’re happy with your new Windows 8.1 boot to VHD, and no longer need the Windows 8 boot to VHD, feel free to delete the old one.  I do believe once you upgrade, you are no longer licensed to use it anyway. There, that was simple wasn’t it? Looking at the huge list of steps it took to perform this upgrade, you may wonder whether I think this is worth it.  Well, I think it is worth booting to VHD.  It makes backups a snap (go to Windows 7, copy the VHD, you backed up the o/s) and helps with disk management – want to move the o/s, you can move the VHD and repoint the boot menu to the new location. The downside is that Microsoft has complete neglected to support boot to VHD as an upgradable option.  Quite a poor decision in my opinion, and if you read twitter and the forums quite a few people agree with that view.  It’s a shame this got missed in the work on creating the upgrade packages for Windows 8.1.

    Read the article

  • The SPARC SuperCluster

    - by Karoly Vegh
    Oracle has been providing a lead in the Engineered Systems business for quite a while now, in accordance with the motto "Hardware and Software Engineered to Work Together." Indeed it is hard to find a better definition of these systems.  Allow me to summarize the idea. It is:  Build a compute platform optimized to run your technologies Develop application aware, intelligently caching storage components Take an impressively fast network technology interconnecting it with the compute nodes Tune the application to scale with the nodes to yet unseen performance Reduce the amount of data moving via compression Provide this all in a pre-integrated single product with a single-pane management interface All these ideas have been around in IT for quite some time now. The real Oracle advantage is adding the last one to put these all together. Oracle has built quite a portfolio of Engineered Systems, to run its technologies - and run those like they never ran before. In this post I'll focus on one of them that serves as a consolidation demigod, a multi-purpose engineered system.  As you probably have guessed, I am talking about the SPARC SuperCluster. It has many great features inherited from its predecessors, and it adds several new ones. Allow me to pick out and elaborate about some of the most interesting ones from a technological point of view.  I. It is the SPARC SuperCluster T4-4. That is, as compute nodes, it includes SPARC T4-4 servers that we learned to appreciate and respect for their features: The SPARC T4 CPUs: Each CPU has 8 cores, each core runs 8 threads. The SPARC T4-4 servers have 4 sockets. That is, a single compute node can in parallel, simultaneously  execute 256 threads. Now, a full-rack SPARC SuperCluster has 4 of these servers on board. Remember the keyword demigod.  While retaining the forerunner SPARC T3's exceptional throughput, the SPARC T4 CPUs raise the bar with single performance too - a humble 5x better one than their ancestors.  actually, the SPARC T4 CPU cores run in both single-threaded and multi-threaded mode, and switch between these two on-the-fly, fulfilling not only single-threaded OR multi-threaded applications' needs, but even mixed requirements (like in database workloads!). Data security, anyone? Every SPARC T4 CPU core has a built-in encryption engine, that is, encryption algorithms cast into silicon.  A PCI controller right on the chip for customers who need I/O performance.  Built-in, no-cost Virtualization:  Oracle VM for SPARC (the former LDoms or Logical Domains) is not a server-emulation virtualization technology but rather a serverpartitioning one, the hypervisor runs in the server firmware, and all the VMs' HW resources (I/O, CPU, memory) are accessed natively, without performance overhead.  This enables customers to run a number of Solaris 10 and Solaris 11 VMs separated, independent of each other within a physical server II. For Database performance, it includes Exadata Storage Cells - one of the main reasons why the Exadata Database Machine performs at diabolic speed. What makes them important? They provide DB backend storage for your Oracle Databases to run on the SPARC SuperCluster, that is what they are built and tuned for DB performance.  These storage cells are SQL-aware.  That is, if a SPARC T4 database compute node executes a query, it doesn't simply request tons of raw datablocks from the storage, filters the received data, and throws away most of it where the statement doesn't apply, but provides the SQL query to the storage node too. The storage cell software speaks SQL, that is, it is able to prefilter and through that transfer only the relevant data. With this, the traffic between database nodes and storage cells is reduced immensely. Less I/O is a good thing - as they say, all the CPUs of the world do one thing just as fast as any other - and that is waiting for I/O.  They don't only pre-filter, but also provide data preprocessing features - e.g. if a DB-node requests an aggregate of data, they can calculate it, and handover only the results, not the whole set. Again, less data to transfer.  They support the magical HCC, (Hybrid Columnar Compression). That is, data can be stored in a precompressed form on the storage. Less data to transfer.  Of course one can't simply rely on disks for performance, there is Flash Storage included there for caching.  III. The low latency, high-speed backbone network: InfiniBand, that interconnects all the members with: Real High Speed: 40 Gbit/s. Full Duplex, of course. Oh, and a really low latency.  RDMA. Remote Direct Memory Access. This technology allows the DB nodes to do exactly that. Remotely, directly placing SQL commands into the Memory of the storage cells. Dodging all the network-stack bottlenecks, avoiding overhead, placing requests directly into the process queue.  You can also run IP over InfiniBand if you please - that's the way the compute nodes can communicate with each other.  IV. Including a general-purpose storage too: the ZFSSA, which is a unified storage, providing NAS and SAN access too, with the following features:  NFS over RDMA over InfiniBand. Nothing is faster network-filesystem-wise.  All the ZFS features onboard, hybrid storage pools, compression, deduplication, snapshot, replication, NFS and CIFS shares Storageheads in a HA-Cluster configuration providing availability of the data  DTrace Live Analytics in a web-based Administration UI Being a general purpose application data storage for your non-database applications running on the SPARC SuperCluster over whichever protocol they prefer, easily replicating, snapshotting, cloning data for them.  There's a lot of great technology included in Oracle's SPARC SuperCluster, we have talked its interior through. As for external scalability: you can start with a half- of full- rack SPARC SuperCluster, and scale out to several racks - that is, stacking not separate full-rack SPARC SuperClusters, but extending always one large instance of the size of several full-racks. Yes, over InfiniBand network. Add racks as you grow.  What technologies shall run on it? SPARC SuperCluster is a general purpose scaleout consolidation/cloud environment. You can run Oracle Databases with RAC scaling, or Oracle Weblogic (end enjoy the SPARC T4's advantages to run Java). Remember, Oracle technologies have been integrated with the Oracle Engineered Systems - this is the Oracle on Oracle advantage. But you can run other software environments such as SAP if you please too. Run any application that runs on Oracle Solaris 10 or Solaris 11. Separate them in Virtual Machines, or even Oracle Solaris Zones, monitor and manage those from a central UI. Here the key takeaways once again: The SPARC SuperCluster: Is a pre-integrated Engineered System Contains SPARC T4-4 servers with built-in virtualization, cryptography, dynamic threading Contains the Exadata storage cells that intelligently offload the burden of the DB-nodes  Contains a highly available ZFS Storage Appliance, that provides SAN/NAS storage in a unified way Combines all these elements over a high-speed, low-latency backbone network implemented with InfiniBand Can grow from a single half-rack to several full-rack size Supports the consolidation of hundreds of applications To summarize: All these technologies are great by themselves, but the real value is like in every other Oracle Engineered System: Integration. All these technologies are tuned to perform together. Together they are way more than the sum of all - and a careful and actually very time consuming integration process is necessary to orchestrate all these for performance. The SPARC SuperCluster's goal is to enable infrastructure operations and offer a pre-integrated solution that can be architected and delivered in hours instead of months of evaluations and tests. The tedious and most importantly time and resource consuming part of the work - testing and evaluating - has been done.  Now go, provide services.   -- charlie  

    Read the article

  • HTG Explains: Should You Build Your Own PC?

    - by Chris Hoffman
    There was a time when every geek seemed to build their own PC. While the masses bought eMachines and Compaqs, geeks built their own more powerful and reliable desktop machines for cheaper. But does this still make sense? Building your own PC still offers as much flexibility in component choice as it ever did, but prebuilt computers are available at extremely competitive prices. Building your own PC will no longer save you money in most cases. The Rise of Laptops It’s impossible to look at the decline of geeks building their own PCs without considering the rise of laptops. There was a time when everyone seemed to use desktops — laptops were more expensive and significantly slower in day-to-day tasks. With the diminishing importance of computing power — nearly every modern computer has more than enough power to surf the web and use typical programs like Microsoft Office without any trouble — and the rise of laptop availability at nearly every price point, most people are buying laptops instead of desktops. And, if you’re buying a laptop, you can’t really build your own. You can’t just buy a laptop case and start plugging components into it — even if you could, you would end up with an extremely bulky device. Ultimately, to consider building your own desktop PC, you have to actually want a desktop PC. Most people are better served by laptops. Benefits to PC Building The two main reasons to build your own PC have been component choice and saving money. Building your own PC allows you to choose all the specific components you want rather than have them chosen for you. You get to choose everything, including the PC’s case and cooling system. Want a huge case with room for a fancy water-cooling system? You probably want to build your own PC. In the past, this often allowed you to save money — you could get better deals by buying the components yourself and combining them, avoiding the PC manufacturer markup. You’d often even end up with better components — you could pick up a more powerful CPU that was easier to overclock and choose more reliable components so you wouldn’t have to put up with an unstable eMachine that crashed every day. PCs you build yourself are also likely more upgradable — a prebuilt PC may have a sealed case and be constructed in such a way to discourage you from tampering with the insides, while swapping components in and out is generally easier with a computer you’ve built on your own. If you want to upgrade your CPU or replace your graphics card, it’s a definite benefit. Downsides to Building Your Own PC It’s important to remember there are downsides to building your own PC, too. For one thing, it’s just more work — sure, if you know what you’re doing, building your own PC isn’t that hard. Even for a geek, researching the best components, price-matching, waiting for them all to arrive, and building the PC just takes longer. Warranty is a more pernicious problem. If you buy a prebuilt PC and it starts malfunctioning, you can contact the computer’s manufacturer and have them deal with it. You don’t need to worry about what’s wrong. If you build your own PC and it starts malfunctioning, you have to diagnose the problem yourself. What’s malfunctioning, the motherboard, CPU, RAM, graphics card, or power supply? Each component has a separate warranty through its manufacturer, so you’ll have to determine which component is malfunctioning before you can send it off for replacement. Should You Still Build Your Own PC? Let’s say you do want a desktop and are willing to consider building your own PC. First, bear in mind that PC manufacturers are buying in bulk and getting a better deal on each component. They also have to pay much less for a Windows license than the $120 or so it would cost you to to buy your own Windows license. This is all going to wipe out the cost savings you’ll see — with everything all told, you’ll probably spend more money building your own average desktop PC than you would picking one up from Amazon or the local electronics store. If you’re an average PC user that uses your desktop for the typical things, there’s no money to be saved from building your own PC. But maybe you’re looking for something higher end. Perhaps you want a high-end gaming PC with the fastest graphics card and CPU available. Perhaps you want to pick out each individual component and choose the exact components for your gaming rig. In this case, building your own PC may be a good option. As you start to look at more expensive, high-end PCs, you may start to see a price gap — but you may not. Let’s say you wanted to blow thousands of dollars on a gaming PC. If you’re looking at spending this kind of money, it would be worth comparing the cost of individual components versus a prebuilt gaming system. Still, the actual prices may surprise you. For example, if you wanted to upgrade Dell’s $2293 Alienware Aurora to include a second NVIDIA GeForce GTX 780 graphics card, you’d pay an additional $600 on Alienware’s website. The same graphics card costs $650 on Amazon or Newegg, so you’d be spending more money building the system yourself. Why? Dell’s Alienware gets bulk discounts you can’t get — and this is Alienware, which was once regarded as selling ridiculously overpriced gaming PCs to people who wouldn’t build their own. Building your own PC still allows you to get the most freedom when choosing and combining components, but this is only valuable to a small niche of gamers and professional users — most people, even average gamers, would be fine going with a prebuilt system. If you’re an average person or even an average gamer, you’ll likely find that it’s cheaper to purchase a prebuilt PC rather than assemble your own. Even at the very high end, components may be more expensive separately than they are in a prebuilt PC. Enthusiasts who want to choose all the individual components for their dream gaming PC and want maximum flexibility may want to build their own PCs. Even then, building your own PC these days is more about flexibility and component choice than it is about saving money. In summary, you probably shouldn’t build your own PC. If you’re an enthusiast, you may want to — but only a small minority of people would actually benefit from building their own systems. Feel free to compare prices, but you may be surprised which is cheaper. Image Credit: Richard Jones on Flickr, elPadawan on Flickr, Richard Jones on Flickr     

    Read the article

  • IBM Server Config questions

    - by Joel Coel
    I have a few questions on a potential server setup. First, the situation: Last year we bought an IBM x3500 server with 2 Xeon E5410's, 9GB RAM, 6 HDDs. The original intent for this server was to replace the old exchange e-mail server. It was brought in, set up, and then shortly after we switched to gmail. Shortly after that my predecessor left for greener pastures, and finally I was hired. So this nice server is now sitting (mostly) idle. This year I have budget again for one server, and of course I want to put this other server to work. I'm thinking about the best use for the two server, and I think I finally have a plan for what I want to do with them. The idea is to use the two newer servers as a pair of VM hosts. I will set up each server with the same 8 VMs, but divide up the load so that only 4 are active per physical host. That means I've normally got 2GB RAM + 2 cores per host. I've done some load testing to pick out what servers to convert to virtual, and chose them so that each host will be capable of handling the entire set of 8 by itself in a pinch with 1 core and 1GB RAM, but would be very taxed to do so. This should take our data center from 13 total servers down to 7. The "servers" I'm replacing are mostly re-purposed desktops, so I'm more than happy to be able to do this. Now it's time to go shopping for the new server. I'd like my two hosts to match as closely as possible, and so I'm looking at IBM again. It also helps that we have some educational matching grant money from IBM that I need to use to help pay for this system (we're a small private college). So finally, (if you're not bored already), we come to my questions: Am I missing anything big or obvious in this plan? I'm a little worried about network performance since the VM hosts will only have 4 nics total where 8 used to be, but I don't think it will be a problem. Is there anything else like this I might be overlooking? Am I making it even too complicated? IBM no longer has a good analog to last year's server. If I want to match the performance (8 cores, 9GB RAM, 1333mhz front side bus, 6 spindles), I have to spend quite a bit more than we paid last year: $2K+, or nearly a 33% cost increase. This only brings a marginal increase in performance. The alternative to stay in budget is to take a hit on the fsb down to 800mhz or cut the number of cores in half, neither of which is attractive. The main cost culprit is the processor. IBM no longer offers the E5410. It's listed as a part, but not available in any of the server configs I've looked at. I'm considering getting the cheapest 800mhz fsb dual core xeon I can configure and then buying the E5410's separately. That's still an extra $350 I wasn't counting on, but that's better than $2K. I want to know what others think of this - will it work or will I end up with the wrong motherboard or some other issue? Am I missing a simple way to configure the server I really want? I don't really intend to do this, but one option to save some money back is to omit the redundant power supply. Since my redundancy plan for these system is to switch over to a completely different host, the extra power isn't fully necessary. That said, it's still very helpful to avoid even short downtimes while I switch over VMs. Has anyone done this?

    Read the article

  • SQL SERVER – Introduction to Wait Stats and Wait Types – Wait Type – Day 1 of 28

    - by pinaldave
    I have been working a lot on Wait Stats and Wait Types recently. Last Year, I requested blog readers to send me their respective server’s wait stats. I appreciate their kind response as I have received  Wait stats from my readers. I took each of the results and carefully analyzed them. I provided necessary feedback to the person who sent me his wait stats and wait types. Based on the feedbacks I got, many of the readers have tuned their server. After a while I got further feedbacks on my recommendations and again, I collected wait stats. I recorded the wait stats and my recommendations and did further research. At some point at time, there were more than 10 different round trips of the recommendations and suggestions. Finally, after six month of working my hands on performance tuning, I have collected some real world wisdom because of this. Now I plan to share my findings with all of you over here. Before anything else, please note that all of these are based on my personal observations and opinions. They may or may not match the theory available at other places. Some of the suggestions may not match your situation. Remember, every server is different and consequently, there is more than one solution to a particular problem. However, this series is written with kept wait stats in mind. While I was working on various performance tuning consultations, I did many more things than just tuning wait stats. Today we will discuss how to capture the wait stats. I use the script diagnostic script created by my friend and SQL Server Expert Glenn Berry to collect wait stats. Here is the script to collect the wait stats: -- Isolate top waits for server instance since last restart or statistics clear WITH Waits AS (SELECT wait_type, wait_time_ms / 1000. AS wait_time_s, 100. * wait_time_ms / SUM(wait_time_ms) OVER() AS pct, ROW_NUMBER() OVER(ORDER BY wait_time_ms DESC) AS rn FROM sys.dm_os_wait_stats WHERE wait_type NOT IN ('CLR_SEMAPHORE','LAZYWRITER_SLEEP','RESOURCE_QUEUE','SLEEP_TASK' ,'SLEEP_SYSTEMTASK','SQLTRACE_BUFFER_FLUSH','WAITFOR', 'LOGMGR_QUEUE','CHECKPOINT_QUEUE' ,'REQUEST_FOR_DEADLOCK_SEARCH','XE_TIMER_EVENT','BROKER_TO_FLUSH','BROKER_TASK_STOP','CLR_MANUAL_EVENT' ,'CLR_AUTO_EVENT','DISPATCHER_QUEUE_SEMAPHORE', 'FT_IFTS_SCHEDULER_IDLE_WAIT' ,'XE_DISPATCHER_WAIT', 'XE_DISPATCHER_JOIN', 'SQLTRACE_INCREMENTAL_FLUSH_SLEEP')) SELECT W1.wait_type, CAST(W1.wait_time_s AS DECIMAL(12, 2)) AS wait_time_s, CAST(W1.pct AS DECIMAL(12, 2)) AS pct, CAST(SUM(W2.pct) AS DECIMAL(12, 2)) AS running_pct FROM Waits AS W1 INNER JOIN Waits AS W2 ON W2.rn <= W1.rn GROUP BY W1.rn, W1.wait_type, W1.wait_time_s, W1.pct HAVING SUM(W2.pct) - W1.pct < 99 OPTION (RECOMPILE); -- percentage threshold GO This script uses Dynamic Management View sys.dm_os_wait_stats to collect the wait stats. It omits the system-related wait stats which are not useful to diagnose performance-related bottleneck. Additionally, not OPTION (RECOMPILE) at the end of the DMV will ensure that every time the query runs, it retrieves new data and not the cached data. This dynamic management view collects all the information since the time when the SQL Server services have been restarted. You can also manually clear the wait stats using the following command: DBCC SQLPERF('sys.dm_os_wait_stats', CLEAR); Once the wait stats are collected, we can start analysis them and try to see what is causing any particular wait stats to achieve higher percentages than the others. Many waits stats are related to one another. When the CPU pressure is high, all the CPU-related wait stats show up on top. But when that is fixed, all the wait stats related to the CPU start showing reasonable percentages. It is difficult to have a sure solution, but there are good indications and good suggestions on how to solve this. I will keep this blog post updated as I will post more details about wait stats and how I reduce them. The reference to Book On Line is over here. Of course, I have selected February to run this Wait Stats series. I am already cheating by having the smallest month to run this series. :) Reference: Pinal Dave (http://blog.SQLAuthority.com) Filed under: DMV, Pinal Dave, PostADay, SQL, SQL Authority, SQL Optimization, SQL Performance, SQL Query, SQL Scripts, SQL Server, SQL Tips and Tricks, SQL Wait Stats, SQL Wait Types, T SQL, Technology

    Read the article

  • Going Parallel with the Task Parallel Library and PLINQ

    With more and more computers using a multi-core processor, the free lunch of increased clock speeds and the inherent performance gains are over. Software developers must instead make sure their applications take use of all the cores available in an efficient manner. New features in .NET 4.0 mean that managed code developers too can join the party.

    Read the article

  • The new Auto Scaling Service in Windows Azure

    - by shiju
    One of the key features of the Cloud is the on-demand scalability, which lets the cloud application developers to scale up or scale down the number of compute resources hosted on the Cloud. Auto Scaling provides the capability to dynamically scale up and scale down your compute resources based on user-defined policies, Key Performance Indicators (KPI), health status checks, and schedules, without any manual intervention. Auto Scaling is an important feature to consider when designing and architecting cloud based solutions, which can unleash the real power of Cloud to the apps for providing truly on-demand scalability and can also guard the organizational budget for cloud based application deployment. In the past, you have had to leverage the the Microsoft Enterprise Library Autoscaling Application Block (WASABi) or a services like  MetricsHub for implementing Automatic Scaling for your cloud apps hosted on the Windows Azure. The WASABi required to host your auto scaling block in a Windows Azure Worker Role for effectively implementing the auto scaling behaviour to your Windows Azure apps. The newly announced Auto Scaling service in Windows Azure lets you add automatic scaling capability to your Windows Azure Compute Services such as Cloud Services, Web Sites and Virtual Machine. Unlike WASABi hosted on a Worker Role, you don’t need to host any monitoring service for using the new Auto Scaling service and the Auto Scaling service will be available to individual Windows Azure Compute Services as part of the Scaling. Configure Auto Scaling for a Windows Azure Cloud Service Currently the Auto Scaling service supports Cloud Services, Web Sites and Virtual Machine. In this demo, I will be used a Cloud Services app with a Web Role and a Worker Role. To enable the Auto Scaling, select t your Windows Azure app in the Windows Azure management portal, and choose “SCLALE” tab. The Scale tab will show the all information regards with Auto Scaling. The below image shows that we have currently disabled the AutoScale service. To enable Auto Scaling, you need to choose either CPU or QUEUE. The QUEUE option is not available for Web Sites. The image below demonstrates how to configure Auto Scaling for a Web Role based on the utilization of CPU. We have configured the web role app for running with 1 to 5 Virtual Machine instances based on the CPU utilization with a range of 50 to 80%. If the aggregate utilization is becoming above above 80%, it will scale up instances and it will scale down instances when utilization is becoming below 50%. The image below demonstrates how to configure Auto Scaling for a Worker Role app based on the messages added into the Windows Azure storage Queue. We configured the worker role app for running with 1 to 3 Virtual Machine instances based on the Queue messages added into the Windows Azure storage Queue. Here we have specified the number of messages target per machine is 2000. The image below shows the summary of the Auto Scaling for the Cloud Service after configuring auto scaling service. Summary Auto Scaling is an extremely important behaviour of the Cloud applications for providing on-demand scalability without any manual intervention. Windows Azure provides greater support for enabling Auto Scaling for the apps deployed on the Windows Azure cloud platform. The new Auto Scaling service in Windows Azure lets you add automatic scaling capability to your Windows Azure Compute Services such as Cloud Services, Web Sites and Virtual Machine. In the new Auto Scaling service, you don’t have to host any monitor service like you have had in WASABi block. The Auto Scaling service is an excellent alternative to the manually hosting WASABi block in a Worker Role app.

    Read the article

  • SQL SERVER – ASYNC_IO_COMPLETION – Wait Type – Day 11 of 28

    - by pinaldave
    For any good system three things are vital: CPU, Memory and IO (disk). Among these three, IO is the most crucial factor of SQL Server. Looking at real-world cases, I do not see IT people upgrading CPU and Memory frequently. However, the disk is often upgraded for either improving the space, speed or throughput. Today we will look at another IO-related wait type. From Book On-Line: Occurs when a task is waiting for I/Os to finish. ASYNC_IO_COMPLETION Explanation: Any tasks are waiting for I/O to finish. If by any means your application that’s connected to SQL Server is processing the data very slowly, this type of wait can occur. Several long-running database operations like BACKUP, CREATE DATABASE, ALTER DATABASE or other operations can also create this wait type. Reducing ASYNC_IO_COMPLETION wait: When it is an issue related to IO, one should check for the following things associated to IO subsystem: Look at the programming and see if there is any application code which processes the data slowly (like inefficient loop, etc.). Note that it should be re-written to avoid this  wait type. Proper placing of the files is very important. We should check the file system for proper placement of the files – LDF and MDF on separate drive, TempDB on another separate drive, hot spot tables on separate filegroup (and on separate disk), etc. Check the File Statistics and see if there is a higher IO Read and IO Write Stall SQL SERVER – Get File Statistics Using fn_virtualfilestats. Check event log and error log for any errors or warnings related to IO. If you are using SAN (Storage Area Network), check the throughput of the SAN system as well as configuration of the HBA Queue Depth. In one of my recent projects, the SAN was performing really badly and so the SAN administrator did not accept it. After some investigations, he agreed to change the HBA Queue Depth on the development setup (test environment). As soon as we changed the HBA Queue Depth to quite a higher value, there was a sudden big improvement in the performance. It is very likely to happen that there are no proper indexes on the system and yet there are lots of table scans and heap scans. Creating proper index can reduce the IO bandwidth considerably. If SQL Server can use appropriate cover index instead of clustered index, it can effectively reduce lots of CPU, Memory and IO (considering cover index has lesser columns than cluster table and all other; it depends upon the situation). You can refer to the following two articles I wrote that talk about how to optimize indexes: Create Missing Indexes Drop Unused Indexes Checking Memory Related Perfmon Counters SQLServer: Memory Manager\Memory Grants Pending (Consistent higher value than 0-2) SQLServer: Memory Manager\Memory Grants Outstanding (Consistent higher value, Benchmark) SQLServer: Buffer Manager\Buffer Hit Cache Ratio (Higher is better, greater than 90% for usually smooth running system) SQLServer: Buffer Manager\Page Life Expectancy (Consistent lower value than 300 seconds) Memory: Available Mbytes (Information only) Memory: Page Faults/sec (Benchmark only) Memory: Pages/sec (Benchmark only) Checking Disk Related Perfmon Counters Average Disk sec/Read (Consistent higher value than 4-8 millisecond is not good) Average Disk sec/Write (Consistent higher value than 4-8 millisecond is not good) Average Disk Read/Write Queue Length (Consistent higher value than benchmark is not good) Read all the post in the Wait Types and Queue series. Note: The information presented here is from my experience and there is no way that I claim it to be accurate. I suggest reading Book OnLine for further clarification. All the discussions of Wait Stats in this blog are generic and vary from system to system. It is recommended that you test this on a development server before implementing it to a production server. Reference: Pinal Dave (http://blog.SQLAuthority.com) Filed under: Pinal Dave, PostADay, SQL, SQL Authority, SQL Query, SQL Scripts, SQL Server, SQL Tips and Tricks, SQL Wait Stats, SQL Wait Types, T SQL, Technology

    Read the article

  • SQL SERVER – IO_COMPLETION – Wait Type – Day 10 of 28

    - by pinaldave
    For any good system three things are vital: CPU, Memory and IO (disk). Among these three, IO is the most crucial factor of SQL Server. Looking at real-world cases, I do not see IT people upgrading CPU and Memory frequently. However, the disk is often upgraded for either improving the space, speed or throughput. Today we will look at an IO-related wait types. From Book On-Line: Occurs while waiting for I/O operations to complete. This wait type generally represents non-data page I/Os. Data page I/O completion waits appear as PAGEIOLATCH_* waits. IO_COMPLETION Explanation: Any tasks are waiting for I/O to finish. This is a good indication that IO needs to be looked over here. Reducing IO_COMPLETION wait: When it is an issue concerning the IO, one should look at the following things related to IO subsystem: Proper placing of the files is very important. We should check the file system for proper placement of files – LDF and MDF on a separate drive, TempDB on another separate drive, hot spot tables on separate filegroup (and on separate disk),etc. Check the File Statistics and see if there is higher IO Read and IO Write Stall SQL SERVER – Get File Statistics Using fn_virtualfilestats. Check event log and error log for any errors or warnings related to IO. If you are using SAN (Storage Area Network), check the throughput of the SAN system as well as the configuration of the HBA Queue Depth. In one of my recent projects, the SAN was performing really badly so the SAN administrator did not accept it. After some investigations, he agreed to change the HBA Queue Depth on development (test environment) set up and as soon as we changed the HBA Queue Depth to quite a higher value, there was a sudden big improvement in the performance. It is very possible that there are no proper indexes in the system and there are lots of table scans and heap scans. Creating proper index can reduce the IO bandwidth considerably. If SQL Server can use appropriate cover index instead of clustered index, it can effectively reduce lots of CPU, Memory and IO (considering cover index has lesser columns than cluster table and all other; it depends upon the situation). You can refer to the two articles that I wrote; they are about how to optimize indexes: Create Missing Indexes Drop Unused Indexes Checking Memory Related Perfmon Counters SQLServer: Memory Manager\Memory Grants Pending (Consistent higher value than 0-2) SQLServer: Memory Manager\Memory Grants Outstanding (Consistent higher value, Benchmark) SQLServer: Buffer Manager\Buffer Hit Cache Ratio (Higher is better, greater than 90% for usually smooth running system) SQLServer: Buffer Manager\Page Life Expectancy (Consistent lower value than 300 seconds) Memory: Available Mbytes (Information only) Memory: Page Faults/sec (Benchmark only) Memory: Pages/sec (Benchmark only) Checking Disk Related Perfmon Counters Average Disk sec/Read (Consistent higher value than 4-8 millisecond is not good) Average Disk sec/Write (Consistent higher value than 4-8 millisecond is not good) Average Disk Read/Write Queue Length (Consistent higher value than benchmark is not good) Note: The information presented here is from my experience and there is no way that I claim it to be accurate. I suggest reading Book OnLine for further clarification. All the discussions of Wait Stats in this blog are generic and vary from system to system. It is recommended that you test this on a development server before implementing it to a production server. Reference: Pinal Dave (http://blog.SQLAuthority.com) Filed under: Pinal Dave, PostADay, SQL, SQL Authority, SQL Query, SQL Scripts, SQL Server, SQL Tips and Tricks, SQL Wait Types, SQL White Papers, T SQL, Technology

    Read the article

  • Monitor System Resources from the Windows 7 Taskbar

    - by Asian Angel
    The problem with most system monitoring apps is that they get covered up with all of your open windows, but you can solve that problem by adding monitoring apps to the Taskbar. Setting Up & Using SuperbarMonitor All of the individual monitors and the .dll files necessary to run them come in a single zip file for your convenience. Simply unzip the contents, add them to an appropriate “Program Files Folder”, and create shortcuts for the monitors that you would like to use on your system. For our example we created shortcuts for all five monitors and set the shortcuts up in their own “Start Menu Folder”. You can see what the five monitors (Battery, CPU, Disk, Memory, & Volume) look like when running…they are visual in appearance without text to clutter up the looks. The monitors use colors (red, green, & yellow) to indicate the amount of resources being used for a particular category. Note: Our system is desktop-based but the “Battery Monitor” was shown for the purposes of demonstration…thus the red color seen here. Hovering the mouse over the “Battery, CPU, Disk, & Memory Monitors” on our system displayed a small blank thumbnail. Note: The “Battery Monitor” may or may not display more when used on your laptop. Going one step further and hovering the mouse over the thumbnails displayed a small blank window. There really is nothing that you will need to worry with outside of watching the color for each individual monitor. Nice and simple! The one monitor with extra features on the thumbnail was the “Volume Monitor”. You can turn the volume down, up, on, or off from here…definitely useful if you have been wanting to hide the “Volume Icon” in the “System Tray”. You can also pin the monitors to your “Taskbar” if desired. Keep in mind that if you do close any of the monitors they will “temporarily” disappear from the “Taskbar” until the next time they are started. Note: If you want the monitors to start with your system each time you will need to add the appropriate shortcuts to the “Startup Sub-menu” in your “Start Menu”. Conclusion If you have been wanting a nice visual way to monitor your system’s resources then SuperbarMonitor is definitely worth trying out. Links Download SuperbarMonitor Similar Articles Productive Geek Tips Monitor CPU, Memory, and Disk IO In Windows 7 with Taskbar MetersUse Windows Vista Reliability Monitor to Troubleshoot CrashesTaskbar Eliminator Does What the Name Implies: Hides Your Windows TaskbarBring Misplaced Off-Screen Windows Back to Your Desktop (Keyboard Trick)How To Fix System Tray Tooltips Not Displaying in Windows XP TouchFreeze Alternative in AutoHotkey The Icy Undertow Desktop Windows Home Server – Backup to LAN The Clear & Clean Desktop Use This Bookmarklet to Easily Get Albums Use AutoHotkey to Assign a Hotkey to a Specific Window Latest Software Reviews Tinyhacker Random Tips DVDFab 6 Revo Uninstaller Pro Registry Mechanic 9 for Windows PC Tools Internet Security Suite 2010 Follow Finder Finds You Twitter Users To Follow Combine MP3 Files Easily QuicklyCode Provides Cheatsheets & Other Programming Stuff Download Free MP3s from Amazon Awe inspiring, inter-galactic theme (Win 7) Case Study – How to Optimize Popular Wordpress Sites

    Read the article

  • World Record Oracle E-Business Consolidated Workload on SPARC T4-2

    - by Brian
    Oracle set a World Record for the Oracle E-Business Suite Standard Medium multiple-online module benchmark using Oracle's SPARC T4-2 and SPARC T4-4 servers which ran the application and database. Oracle's SPARC T4 servers demonstrate performance leadership and world-record results on Oracle E-Business Suite Applications R12 OLTP benchmark by publishing the first result using multiple concurrent online application modules with Oracle Database 11g Release 2 running Solaris.   This results shows that a multi-tier configuration of SPARC T4 servers running the Oracle E-Business Suite R12.1.2 application and Oracle Database 11g Release 2 is capable of supporting 4,100 online users with outstanding response-times, executing a mix of complex transactions consolidating 4 Oracle E-Business modules (iProcurement, Order Management, Customer Service and HR Self-Service).   The SPARC T4-2 server in the application tier utilized about 65% and the SPARC T4-4 server in the database tier utilized about 30%, providing significant headroom for additional Oracle E-Business Suite R12.1.2 processing modules, more online users, and future growth.   Oracle E-Business Suite Applications were run in Oracle Solaris Containers on SPARC T4 servers and provides a consolidation platform for multiple E-Business instances.   Performance Landscape Multiple Online Modules (Self-Service, Order-Management, iProcurement, Customer-Service) Medium Configuration System Users AverageResponse Time 90th PercentileResponse Time SPARC T4-2 4,100 2.08 sec 2.52 sec Configuration Summary Application Tier Configuration: 1 x SPARC T4-2 server 2 x SPARC T4 processors, 2.85 GHz 256 GB memory 3 x 300 GB internal disks Oracle Solaris 10 Oracle E-Business Suite 12.1.2 Database Tier Configuration: 1 x SPARC T4-4 server 4 x SPARC T4 processors, 3.0 GHz 256 GB memory 2 x 300 GB internal disks Oracle Solaris 10 Oracle Solaris Containers Oracle Database 11g Release 2 Storage Configuration: 1 x Sun Storage F5100 Flash Array (80 x 24 GB flash modules) Benchmark Description The Oracle R12 E-Business Suite Standard Benchmark combines online transaction execution by simulated users with multiple online concurrent modules to model a typical scenario for a global enterprise. The online component exercises the common UI flows which are most frequently used by a majority of our customers. This benchmark utilized four concurrent flows of OLTP transactions, for Order to Cash, iProcurement, Customer Service and HR Self-Service and measured the response times. The selected flows model simultaneous business activities inclusive of managing customers, services, products and employees. See Also Oracle R12 E-Business Suite Standard Benchmark Results Oracle R12 E-Business Suite Standard Benchmark Overview Oracle R12 E-Business Benchmark Description E-Business Suite Applications R2 (R12.1.2) Online Benchmark - Using Oracle Database 11g on Oracle's SPARC T4-2 and Oracle's SPARC T4-4 Servers oracle.com SPARC T4-2 Server oracle.com OTN SPARC T4-4 Server oracle.com OTN Oracle E-Business Suite oracle.com OTN Oracle Solaris oracle.com OTN Oracle Database 11g Release 2 Enterprise Edition oracle.com OTN Disclosure Statement Oracle E-Business Suite R12 medium multiple-online module benchmark, SPARC T4-2, SPARC T4, 2.85 GHz, 2 chips, 16 cores, 128 threads, 256 GB memory, SPARC T4-4, SPARC T4, 3.0 GHz, 4 chips, 32 cores, 256 threads, 256 GB memory, average response time 2.08 sec, 90th percentile response time 2.52 sec, Oracle Solaris 10, Oracle Solaris Containers, Oracle E-Business Suite 12.1.2, Oracle Database 11g Release 2, Results as of 9/30/2012.

    Read the article

  • Como estão os seus projetos em TI? ALM (Application lifecycle management) - Parte 1

    - by johnywercley
    O gráfico mostra um número assustador, em outras palavras, no mundo inteiro as coisas não andam bem, são pesquisas feitas por um importante orgão o “Stand Group”. Eles nos chamam atenção a quantidade de projetos com problemas, fazendo uma análise primária, somando a parte verde com azul veremos a porcentagem de projetos TI com problemas, projetos que chegam a de fato dar certo, são os de cores vermelhas, um número muito baixo. Se você fosse hoje investidor financeiro e tivesse que fazer um projeto...(read more)

    Read the article

  • IPgallery banks on Solaris SPARC

    - by Frederic Pariente
    IPgallery is a global supplier of converged legacy and Next Generation Networks (NGN) products and solutions, including: core network components and cloud-based Value Added Services (VAS) for voice, video and data sessions. IPgallery enables network operators and service providers to offer advanced converged voice, chat, video/content services and rich unified social communications in a combined legacy (fixed/mobile), Over-the-Top (OTT) and Social Community (SC) environments for home and business customers. Technically speaking, this offer is a scalable and robust telco solution enabling operators to offer new services while controlling operating expenses (OPEX). In its solutions, IPgallery leverages the following Oracle components: Oracle Solaris, Netra T4 and SPARC T4 in order to provide a competitive and scalable solution without the price tag often associated with high-end systems. Oracle Solaris Binary Application Guarantee A unique feature of Oracle Solaris is the guaranteed binary compatibility between releases of the Solaris OS. That means, if a binary application runs on Solaris 2.6 or later, it will run on the latest release of Oracle Solaris.  IPgallery developed their application on Solaris 9 and Solaris 10 then runs it on Solaris 11, without any code modification or rebuild. The Solaris Binary Application Guarantee helps IPgallery protect their long-term investment in the development, training and maintenance of their applications. Oracle Solaris Image Packaging System (IPS) IPS is a new repository-based package management system that comes with Oracle Solaris 11. It provides a framework for complete software life-cycle management such as installation, upgrade and removal of software packages. IPgallery leverages this new packaging system in order to speed up and simplify software installation for the R&D and production environments. Notably, they use IPS to deliver Solaris Studio 12.3 packages as part of the rapid installation process of R&D environments, and during the production software deployment phase, they ensure software package integrity using the built-in verification feature. Solaris IPS thus improves IPgallery's time-to-market with a faster, more reliable software installation and deployment in production environments. Extreme Network Performance IPgallery saw a huge improvement in application performance both in CPU and I/O, when running on SPARC T4 architecture in compared to UltraSPARC T2 servers.  The same application (with the same activation environment) running on T2 consumes 40%-50% CPU, while it consumes only 10% of the CPU on T4. The testing environment comprised of: Softswitch (Call management), TappS (Telecom Application Server) and Billing Server running on same machine and initiating various services in capacity of 1000 CAPS (Call Attempts Per Second). In addition, tests showed a huge improvement in the performance of the TCP/IP stack, which reduces network layer processing and in the end Call Attempts latency. Finally, there is a huge improvement within the file system and disk I/O operations; they ran all tests with maximum logging capability and it didn't influence any benchmark values. "Due to the huge improvements in performance and capacity using the T4-1 architecture, IPgallery has engineered the solution with less hardware.  This means instead of deploying the solution on six T2-based machines, we will deploy on 2 redundant machines while utilizing Oracle Solaris Zones and Oracle VM for higher availability and virtualization" Shimon Lichter, VP R&D, IPgallery In conclusion, using the unique combination of Oracle Solaris and SPARC technologies, IPgallery is able to offer solutions with much lower TCO, while providing a higher level of service capacity, scalability and resiliency. This low-OPEX solution enables the operator, the end-customer, to deliver a high quality service while maintaining high profitability.

    Read the article

  • Why don't xUnit frameworks allow tests to run in parallel?

    - by Xavier Nodet
    Do you know of any xUnit framework that allows to run tests in parallel, to make use of multiple cores in today's machine? I don't... If none (or so few) of them does it, maybe there is a reason... Is it that tests are usually so quick that people simply don't feel the need to paralellize them? Is there something deeper that precludes distributing (at least some of) the tests over multiple threads? Thanks!

    Read the article

  • Slow Ubuntu 10.04 after long time unused

    - by Winston Ewert
    I'm at spring break so I'm back at my parent's house. I've turned my computer on which has been off since January and its unusably slow. This was not the case when I last used the computer in January. It is running 10.04, Memory: 875.5 MB CPU: AMD Athlon 64 X2 Dual Core Processor 4400+ Available Disk Space: 330.8 GB I'm not seeing a large usage of either memory or Disk I/O. If I look at my list of processes there is only a very small amount of CPU usage. However, if I hover over the CPU usage graph that I've on the top bar, I sometimes get really high readings like 100%. It took a long time to boot, to open firefox, to open a link in firefox. As far as I can tell everything that the computer tries to do is just massively slow. Right now, I'm apt-get dist-upgrading to install any updates that I will have missed since last time this computer was on. Any ideas as to what is going on here? UPDATE: I thought to check dmesg and it has a lot of entries like this: [ 1870.142201] ata3.00: exception Emask 0x0 SAct 0x7 SErr 0x0 action 0x0 [ 1870.142206] ata3.00: irq_stat 0x40000008 [ 1870.142210] ata3.00: failed command: READ FPDMA QUEUED [ 1870.142217] ata3.00: cmd 60/08:10:c0:4a:65/00:00:03:00:00/40 tag 2 ncq 4096 in [ 1870.142218] res 41/40:00:c5:4a:65/00:00:03:00:00/40 Emask 0x409 (media error) <F> [ 1870.142221] ata3.00: status: { DRDY ERR } [ 1870.142223] ata3.00: error: { UNC } [ 1870.143981] ata3.00: SB600 AHCI: limiting to 255 sectors per cmd [ 1870.146758] ata3.00: SB600 AHCI: limiting to 255 sectors per cmd [ 1870.146761] ata3.00: configured for UDMA/133 [ 1870.146777] ata3: EH complete [ 1872.092269] ata3.00: exception Emask 0x0 SAct 0x7 SErr 0x0 action 0x0 [ 1872.092274] ata3.00: irq_stat 0x40000008 [ 1872.092278] ata3.00: failed command: READ FPDMA QUEUED [ 1872.092285] ata3.00: cmd 60/08:00:c0:4a:65/00:00:03:00:00/40 tag 0 ncq 4096 in [ 1872.092287] res 41/40:00:c5:4a:65/00:00:03:00:00/40 Emask 0x409 (media error) <F> [ 1872.092289] ata3.00: status: { DRDY ERR } [ 1872.092292] ata3.00: error: { UNC } [ 1872.094050] ata3.00: SB600 AHCI: limiting to 255 sectors per cmd [ 1872.096795] ata3.00: SB600 AHCI: limiting to 255 sectors per cmd [ 1872.096798] ata3.00: configured for UDMA/133 [ 1872.096814] ata3: EH complete [ 1874.042279] ata3.00: exception Emask 0x0 SAct 0x7 SErr 0x0 action 0x0 [ 1874.042285] ata3.00: irq_stat 0x40000008 [ 1874.042289] ata3.00: failed command: READ FPDMA QUEUED [ 1874.042296] ata3.00: cmd 60/08:10:c0:4a:65/00:00:03:00:00/40 tag 2 ncq 4096 in [ 1874.042297] res 41/40:00:c5:4a:65/00:00:03:00:00/40 Emask 0x409 (media error) <F> [ 1874.042300] ata3.00: status: { DRDY ERR } [ 1874.042302] ata3.00: error: { UNC } [ 1874.044048] ata3.00: SB600 AHCI: limiting to 255 sectors per cmd [ 1874.046837] ata3.00: SB600 AHCI: limiting to 255 sectors per cmd [ 1874.046840] ata3.00: configured for UDMA/133 [ 1874.046861] sd 2:0:0:0: [sda] Unhandled sense code [ 1874.046863] sd 2:0:0:0: [sda] Result: hostbyte=DID_OK driverbyte=DRIVER_SENSE [ 1874.046867] sd 2:0:0:0: [sda] Sense Key : Medium Error [current] [descriptor] [ 1874.046872] Descriptor sense data with sense descriptors (in hex): [ 1874.046874] 72 03 11 04 00 00 00 0c 00 0a 80 00 00 00 00 00 [ 1874.046883] 03 65 4a c5 [ 1874.046886] sd 2:0:0:0: [sda] Add. Sense: Unrecovered read error - auto reallocate failed [ 1874.046892] sd 2:0:0:0: [sda] CDB: Read(10): 28 00 03 65 4a c0 00 00 08 00 [ 1874.046900] end_request: I/O error, dev sda, sector 56969925 [ 1874.046920] ata3: EH complete I'm not certain, but that looks like my problem may be a failing hard drive. But the drive is less then a year old, it really shouldn't be failing now...

    Read the article

  • Windows Azure Role Instance Limits

    - by kaleidoscope
    Brief overview of the limits imposed on hosted services in Windows Azure is as follows: Effective before Dec. 10th 2009 Effective  after Dec. 10th 2009 Effective after Jan. 4th 2010 Token (CTP) Token (CTP) Token (non-billing country) Paying subscription Deployment Slots 2 2 2 2 Hosted Services 1 1 20 20 Roles per  deployment 5 5 5 5 Instances per Role 2 2 no limit no limit VM CPU Cores no limit 8 8 20 Storage Accounts 2 2 5 5 More Information: http://blog.toddysm.com/2010/01/windows-azure-role-instance-limits-explained.html   Amit, S

    Read the article

  • NUMA-aware placement of communication variables

    - by Dave
    For classic NUMA-aware programming I'm typically most concerned about simple cold, capacity and compulsory misses and whether we can satisfy the miss by locally connected memory or whether we have to pull the line from its home node over the coherent interconnect -- we'd like to minimize channel contention and conserve interconnect bandwidth. That is, for this style of programming we're quite aware of where memory is homed relative to the threads that will be accessing it. Ideally, a page is collocated on the node with the thread that's expected to most frequently access the page, as simple misses on the page can be satisfied without resorting to transferring the line over the interconnect. The default "first touch" NUMA page placement policy tends to work reasonable well in this regard. When a virtual page is first accessed, the operating system will attempt to provision and map that virtual page to a physical page allocated from the node where the accessing thread is running. It's worth noting that the node-level memory interleaving granularity is usually a multiple of the page size, so we can say that a given page P resides on some node N. That is, the memory underlying a page resides on just one node. But when thinking about accesses to heavily-written communication variables we normally consider what caches the lines underlying such variables might be resident in, and in what states. We want to minimize coherence misses and cache probe activity and interconnect traffic in general. I don't usually give much thought to the location of the home NUMA node underlying such highly shared variables. On a SPARC T5440, for instance, which consists of 4 T2+ processors connected by a central coherence hub, the home node and placement of heavily accessed communication variables has very little impact on performance. The variables are frequently accessed so likely in M-state in some cache, and the location of the home node is of little consequence because a requester can use cache-to-cache transfers to get the line. Or at least that's what I thought. Recently, though, I was exploring a simple shared memory point-to-point communication model where a client writes a request into a request mailbox and then busy-waits on a response variable. It's a simple example of delegation based on message passing. The server polls the request mailbox, and having fetched a new request value, performs some operation and then writes a reply value into the response variable. As noted above, on a T5440 performance is insensitive to the placement of the communication variables -- the request and response mailbox words. But on a Sun/Oracle X4800 I noticed that was not the case and that NUMA placement of the communication variables was actually quite important. For background an X4800 system consists of 8 Intel X7560 Xeons . Each package (socket) has 8 cores with 2 contexts per core, so the system is 8x8x2. Each package is also a NUMA node and has locally attached memory. Every package has 3 point-to-point QPI links for cache coherence, and the system is configured with a twisted ladder "mobius" topology. The cache coherence fabric is glueless -- there's not central arbiter or coherence hub. The maximum distance between any two nodes is just 2 hops over the QPI links. For any given node, 3 other nodes are 1 hop distant and the remaining 4 nodes are 2 hops distant. Using a single request (client) thread and a single response (server) thread, a benchmark harness explored all permutations of NUMA placement for the two threads and the two communication variables, measuring the average round-trip-time and throughput rate between the client and server. In this benchmark the server simply acts as a simple transponder, writing the request value plus 1 back into the reply field, so there's no particular computation phase and we're only measuring communication overheads. In addition to varying the placement of communication variables over pairs of nodes, we also explored variations where both variables were placed on one page (and thus on one node) -- either on the same cache line or different cache lines -- while varying the node where the variables reside along with the placement of the threads. The key observation was that if the client and server threads were on different nodes, then the best placement of variables was to have the request variable (written by the client and read by the server) reside on the same node as the client thread, and to place the response variable (written by the server and read by the client) on the same node as the server. That is, if you have a variable that's to be written by one thread and read by another, it should be homed with the writer thread. For our simple client-server model that means using split request and response communication variables with unidirectional message flow on a given page. This can yield up to twice the throughput of less favorable placement strategies. Our X4800 uses the QPI 1.0 protocol with source-based snooping. Briefly, when node A needs to probe a cache line it fires off snoop requests to all the nodes in the system. Those recipients then forward their response not to the original requester, but to the home node H of the cache line. H waits for and collects the responses, adjudicates and resolves conflicts and ensures memory-model ordering, and then sends a definitive reply back to the original requester A. If some node B needed to transfer the line to A, it will do so by cache-to-cache transfer and let H know about the disposition of the cache line. A needs to wait for the authoritative response from H. So if a thread on node A wants to write a value to be read by a thread on node B, the latency is dependent on the distances between A, B, and H. We observe the best performance when the written-to variable is co-homed with the writer A. That is, we want H and A to be the same node, as the writer doesn't need the home to respond over the QPI link, as the writer and the home reside on the very same node. With architecturally informed placement of communication variables we eliminate at least one QPI hop from the critical path. Newer Intel processors use the QPI 1.1 coherence protocol with home-based snooping. As noted above, under source-snooping a requester broadcasts snoop requests to all nodes. Those nodes send their response to the home node of the location, which provides memory ordering, reconciles conflicts, etc., and then posts a definitive reply to the requester. In home-based snooping the snoop probe goes directly to the home node and are not broadcast. The home node can consult snoop filters -- if present -- and send out requests to retrieve the line if necessary. The 3rd party owner of the line, if any, can respond either to the home or the original requester (or even to both) according to the protocol policies. There are myriad variations that have been implemented, and unfortunately vendor terminology doesn't always agree between vendors or with the academic taxonomy papers. The key is that home-snooping enables the use of a snoop filter to reduce interconnect traffic. And while home-snooping might have a longer critical path (latency) than source-based snooping, it also may require fewer messages and less overall bandwidth. It'll be interesting to reprise these experiments on a platform with home-based snooping. While collecting data I also noticed that there are placement concerns even in the seemingly trivial case when both threads and both variables reside on a single node. Internally, the cores on each X7560 package are connected by an internal ring. (Actually there are multiple contra-rotating rings). And the last-level on-chip cache (LLC) is partitioned in banks or slices, which with each slice being associated with a core on the ring topology. A hardware hash function associates each physical address with a specific home bank. Thus we face distance and topology concerns even for intra-package communications, although the latencies are not nearly the magnitude we see inter-package. I've not seen such communication distance artifacts on the T2+, where the cache banks are connected to the cores via a high-speed crossbar instead of a ring -- communication latencies seem more regular.

    Read the article

  • Columnstore Case Study #1: MSIT SONAR Aggregations

    - by aspiringgeek
    Preamble This is the first in a series of posts documenting big wins encountered using columnstore indexes in SQL Server 2012 & 2014.  Many of these can be found in this deck along with details such as internals, best practices, caveats, etc.  The purpose of sharing the case studies in this context is to provide an easy-to-consume quick-reference alternative. Why Columnstore? If we’re looking for a subset of columns from one or a few rows, given the right indexes, SQL Server can do a superlative job of providing an answer. If we’re asking a question which by design needs to hit lots of rows—DW, reporting, aggregations, grouping, scans, etc., SQL Server has never had a good mechanism—until columnstore. Columnstore indexes were introduced in SQL Server 2012. However, they're still largely unknown. Some adoption blockers existed; yet columnstore was nonetheless a game changer for many apps.  In SQL Server 2014, potential blockers have been largely removed & they're going to profoundly change the way we interact with our data.  The purpose of this series is to share the performance benefits of columnstore & documenting columnstore is a compelling reason to upgrade to SQL Server 2014. App: MSIT SONAR Aggregations At MSIT, performance & configuration data is captured by SCOM. We archive much of the data in a partitioned data warehouse table in SQL Server 2012 for reporting via an application called SONAR.  By definition, this is a primary use case for columnstore—report queries requiring aggregation over large numbers of rows.  New data is refreshed each night by an automated table partitioning mechanism—a best practices scenario for columnstore. The Win Compared to performance using classic indexing which resulted in the expected query plan selection including partition elimination vs. SQL Server 2012 nonclustered columnstore, query performance increased significantly.  Logical reads were reduced by over a factor of 50; both CPU & duration improved by factors of 20 or more.  Other than creating the columnstore index, no special modifications or tweaks to the app or databases schema were necessary to achieve the performance improvements.  Existing nonclustered indexes were rendered superfluous & were deleted, thus mitigating maintenance challenges such as defragging as well as conserving disk capacity. Details The table provides the raw data & summarizes the performance deltas. Logical Reads (8K pages) CPU (ms) Durn (ms) Columnstore 160,323 20,360 9,786 Conventional Table & Indexes 9,053,423 549,608 193,903 ? x56 x27 x20 The charts provide additional perspective of this data.  "Conventional vs. Columnstore Metrics" document the raw data.  Note on this linear display the magnitude of the conventional index performance vs. columnstore.  The “Metrics (?)” chart expresses these values as a ratio. Summary For DW, reports, & other BI workloads, columnstore often provides significant performance enhancements relative to conventional indexing.  I have documented here, the first in a series of reports on columnstore implementations, results from an initial implementation at MSIT in which logical reads were reduced by over a factor of 50; both CPU & duration improved by factors of 20 or more.  Subsequent features in this series document performance enhancements that are even more significant. 

    Read the article

  • Looking to trade a 1U HP Proliant DL360 G5 in exchange for a small linux VPS

    - by user597875
    I have a 1U HP Proliant DL360 G5 that I have no place to rack and would like to trade it for a small linux VPS. If interested let me know... Here are the specs of the server: Model: Intel Xeon CPU 5150 @ 2.66GHz, 4MB L2 Cache Processor Speed: 2.7GHz Processor Sockets: 2 Processor Cores per Socket: 2 Logical Processors: 4 8GB of memory 4x72GB 10k SAS drives Manufacturer: HP Model: Proliant DL360 G5 BIOS Version: P58

    Read the article

  • Good DBAs Do Baselines

    - by Louis Davidson
    One morning, you wake up and feel funny. You can’t quite put your finger on it, but something isn’t quite right. What now? Unless you happen to be a hypochondriac, you likely drag yourself out of bed, get on with the day and gather more “evidence”. You check your symptoms over the next few days; do you feel the same, better, worse? If better, then great, it was some temporal issue, perhaps caused by an allergic reaction to some suspiciously spicy chicken. If the same or worse then you go to the doctor for some health advice, but armed with some data to share, and having ruled out certain possible causes that are fixed with a bit of rest and perhaps an antacid. Whether you realize it or not, in comparing how you feel one day to the next, you have taken baseline measurements. In much the same way, a DBA uses baselines to gauge the gauge health of their database servers. Of course, while SQL Server is very willing to share data regarding its health and activities, it has almost no idea of the difference between good and bad. Over time, experienced DBAs develop “mental” baselines with which they can gauge the health of their servers almost as easily as their own body. They accumulate knowledge of the daily, natural state of each part of their database system, and so know instinctively when one of their databases “feels funny”. Equally, they know when an “issue” is just a passing tremor. They see their SQL Server with all of its four CPU cores running close 100% and don’t panic anymore. Why? It’s 5PM and every day the same thing occurs when the end-of-day reports, which are very CPU intensive, are running. Equally, they know when they need to respond in earnest when it is the first time they have heard about an issue, even if it has been happening every day. Nevertheless, no DBA can retain mental baselines for every characteristic of their systems, so we need to collect physical baselines too. In my experience, surprisingly few DBAs do this very well. Part of the problem is that SQL Server provides a lot of instrumentation. If you look, you will find an almost overwhelming amount of data regarding user activity on your SQL Server instances, and use and abuse of the available CPU, I/O and memory. It seems like a huge task even to work out which data you need to collect, let alone start collecting it on a regular basis, managing its storage over time, and performing detailed comparative analysis. However, without baselines, though, it is very difficult to pinpoint what ails a server, just by looking at a single snapshot of the data, or to spot retrospectively what caused the problem by examining aggregated data for the server, collected over many months. It isn’t as hard as you think to get started. You’ve probably already established some troubleshooting queries of the type SELECT Value FROM SomeSystemTableOrView. Capturing a set of baseline values for such a query can be as easy as changing it as follows: INSERT into BaseLine.SomeSystemTable (value, captureTime) SELECT Value, SYSDATETIME() FROM SomeSystemTableOrView; Of course, there are monitoring tools that will collect and manage this baseline data for you, automatically, and allow you to perform comparison of metrics over different periods. However, to get yourself started and to prove to yourself (or perhaps the person who writes the checks for tools) the value of baselines, stick something similar to the above query into an agent job, running every hour or so, and you are on your way with no excuses! Then, the next time you investigate a slow server, and see x open transactions, y users logged in, and z rows added per hour in the Orders table, compare to your baselines and see immediately what, if anything, has changed!

    Read the article

< Previous Page | 73 74 75 76 77 78 79 80 81 82 83 84  | Next Page >