Search Results

Search found 267 results on 11 pages for 'benchmarking'.

Page 10/11 | < Previous Page | 6 7 8 9 10 11  | Next Page >

• g-wan - reproducing the performance claims

  - by user2603628

  Using gwan_linux64-bit.tar.bz2 under Ubuntu 12.04 LTS unpacking and running gwan then pointing wrk at it (using a null file null.html) wrk --timeout 10 -t 2 -c 100 -d20s http://127.0.0.1:8080/null.html Running 20s test @ http://127.0.0.1:8080/null.html 2 threads and 100 connections Thread Stats Avg Stdev Max +/- Stdev Latency 11.65s 5.10s 13.89s 83.91% Req/Sec 3.33k 3.65k 12.33k 75.19% 125067 requests in 20.01s, 32.08MB read Socket errors: connect 0, read 37, write 0, timeout 49 Requests/sec: 6251.46 Transfer/sec: 1.60MB .. very poor performance, in fact there seems to be some kind of huge latency issue. During the test gwan is 200% busy and wrk is 67% busy. Pointing at nginx, wrk is 200% busy and nginx is 45% busy: wrk --timeout 10 -t 2 -c 100 -d20s http://127.0.0.1/null.html Thread Stats Avg Stdev Max +/- Stdev Latency 371.81us 134.05us 24.04ms 91.26% Req/Sec 72.75k 7.38k 109.22k 68.21% 2740883 requests in 20.00s, 540.95MB read Requests/sec: 137046.70 Transfer/sec: 27.05MB Pointing weighttpd at nginx gives even faster results: /usr/local/bin/weighttp -k -n 2000000 -c 500 -t 3 http://127.0.0.1/null.html weighttp - a lightweight and simple webserver benchmarking tool starting benchmark... spawning thread #1: 167 concurrent requests, 666667 total requests spawning thread #2: 167 concurrent requests, 666667 total requests spawning thread #3: 166 concurrent requests, 666666 total requests progress: 9% done progress: 19% done progress: 29% done progress: 39% done progress: 49% done progress: 59% done progress: 69% done progress: 79% done progress: 89% done progress: 99% done finished in 7 sec, 13 millisec and 293 microsec, 285172 req/s, 57633 kbyte/s requests: 2000000 total, 2000000 started, 2000000 done, 2000000 succeeded, 0 failed, 0 errored status codes: 2000000 2xx, 0 3xx, 0 4xx, 0 5xx traffic: 413901205 bytes total, 413901205 bytes http, 0 bytes data The server is a virtual 8 core dedicated server (bare metal), under KVM Where do I start looking to identify the problem gwan is having on this platform ? I have tested lighttpd, nginx and node.js on this same OS, and the results are all as one would expect. The server has been tuned in the usual way with expanded ephemeral ports, increased ulimits, adjusted time wait recycling etc.

  Read the article

• Guidance required: FIrst time gonna work with real high end database (size = 50GB).

  - by claws

  I got a project of designing a Database. This is going to be my first big scale project. Good thing about it is information is mostly organized & currently stored in text files. The size of this information is 50GB. There are going to be few millions of records in each Table. Its going to have around 50 tables. I need to provide a web interface for searching & browsing. I'm going to use MySQL DBMS. I've never worked with a database more than 200MB before. So, speed & performance was never a concern but I followed things like normalization & Indexes. I never used any kind of testing/benchmarking/queryOptimization/whatever because I never had to care about them. But here the purpose of creating a database is to make it quickly searchable. So, I need to consider all possible aspects in design. I was browsing archives & found: http://stackoverflow.com/questions/1981526/what-should-every-developer-know-about-databases http://stackoverflow.com/questions/621884/database-development-mistakes-made-by-app-developers I'm gonna keep the points mentioned in above answers in mind. What else should I know? What else should I keep in mind?

  Read the article

• Batch insert mode with hibernate and oracle: seems to be dropping back to slow mode silently

  - by Chris

  I'm trying to get a batch insert working with Hibernate into Oracle, according to what i've read here: http://docs.jboss.org/hibernate/core/3.3/reference/en/html/batch.html , but with my benchmarking it doesn't seem any faster than before. Can anyone suggest a way to prove whether hibernate is using batch mode or not? I hear that there are numerous reasons why it may silently drop into normal mode (eg associations and generated ids) so is there some way to find out why it has gone non-batch? My hibernate.cfg.xml contains this line which i believe is all i need to enable batch mode: <property name="jdbc.batch_size">50</property> My insert code looks like this: List<LogEntry> entries = ..a list of 100 LogEntry data classes... Session sess = sessionFactory.getCurrentSession(); for(LogEntry e : entries) { sess.save(e); } sess.flush(); sess.clear(); My 'logentry' class has no associations, the only interesting field is the id: @Entity @Table(name="log_entries") public class LogEntry { @Id @GeneratedValue public Long id; ..other fields - strings and ints... However, since it is oracle, i believe the @GeneratedValue will use the sequence generator. And i believe that only the 'identity' generator will stop bulk inserts. So if anyone can explain why it isn't running in batch mode, or how i can find out for sure if it is or isn't in batch mode, or find out why hibernate is silently dropping back to slow mode, i'd be most grateful. Thanks

  Read the article

• APC decreasing php performance??? (php 5.3, apache 2.2, windows vista 64bit)

  - by M.M.

  Hi, I have an Apache/2.2.15 (VC9) and PHP/5.3.2 (VC9 thread safe) running as an apache module on Vista 64bit machine. All running fine. Project that I'm benchmarking (with apache's ab utility) is basically standard Zend Framework project with no db connection involved. Average (median) apache response is about 0.15 seconds. After I've installed APC (3.1.4-dev VC9 thread safe) with standard settings suddenly the request response time raised to 1.3 seconds (!), which is unacceptable... All apc settings looked always good (through the apc.php script: enough shm memory, no cache full, fragmentation 0%). Only difference was to disable the stats lookup (apc.stat = 0). Then the response dropped to 0.09 seconds which was finally better than without the apc. IIRC, it's expected and obvious that the stat lookup creates some overhead, but shouldn't it still be far more performant compared to running wihout the apc extension at all? Or put it differently why is the apc.stat creating so much overhead? Apparently, something is not working as it should, I don't really know where to start looking. Thank you for your time/answers/direction in advance. Cheers, m.

  Read the article

• What is the fastest (to access) struct-like object in Python?

  - by DNS

  I'm optimizing some code whose main bottleneck is running through and accessing a very large list of struct-like objects. Currently I'm using namedtuples, for readability. But some quick benchmarking using 'timeit' shows that this is really the wrong way to go where performance is a factor: Named tuple with a, b, c: >>> timeit("z = a.c", "from __main__ import a") 0.38655471766332994 Class using __slots__, with a, b, c: >>> timeit("z = b.c", "from __main__ import b") 0.14527461047146062 Dictionary with keys a, b, c: >>> timeit("z = c['c']", "from __main__ import c") 0.11588272541098377 Tuple with three values, using a constant key: >>> timeit("z = d[2]", "from __main__ import d") 0.11106188992948773 List with three values, using a constant key: >>> timeit("z = e[2]", "from __main__ import e") 0.086038238242508669 Tuple with three values, using a local key: >>> timeit("z = d[key]", "from __main__ import d, key") 0.11187358437882722 List with three values, using a local key: >>> timeit("z = e[key]", "from __main__ import e, key") 0.088604143037173344 First of all, is there anything about these little timeit tests that would render them invalid? I ran each several times, to make sure no random system event had thrown them off, and the results were almost identical. It would appear that dictionaries offer the best balance between performance and readability, with classes coming in second. This is unfortunate, since, for my purposes, I also need the object to be sequence-like; hence my choice of namedtuple. Lists are substantially faster, but constant keys are unmaintainable; I'd have to create a bunch of index-constants, i.e. KEY_1 = 1, KEY_2 = 2, etc. which is also not ideal. Am I stuck with these choices, or is there an alternative that I've missed?

  Read the article

• How do I choose what and when to cache data with ob_start rather than query the database?

  - by Tim Santeford

  I have a home page that has several independent dynamic parts. The parts consist of a list of recent news from the company, a site statistics panel, and the online status of certain employees. The recent news changes monthly, site statistics change daily, and online statuses change on a per minute bases. I would like to cache these panels so that the db is not hit on every page load. Is using ob_start() then ob_get_contents() to cache these parts to a file the correct way to do this or is there a better method in PHP5 for doing this? In asking this question I'm trying to answer these additional questions: How can I determine the correct approach for caching this data without doing extensive benchmarking? Does it make sense to cache these parts in different files and then join them together per requests or should I re-query the data and cache once per minute? I'm looking for a rule of thumb for planning pages and for situations where doing testing is not cost effective (The client is not paying enough for it I mean). Thanks!

  Read the article

• F# why my recursion is faster than Seq.exists?

  - by user38397

  I am pretty new to F#. I'm trying to understand how I can get a fast code in F#. For this, I tried to write two methods (IsPrime1 and IsPrime2) for benchmarking. My code is: // Learn more about F# at http://fsharp.net open System open System.Diagnostics #light let isDivisible n d = n % d = 0 let IsPrime1 n = Array.init (n-2) ((+) 2) |> Array.exists (isDivisible n) |> not let rec hasDivisor n d = match d with | x when x < n -> (n % x = 0) || (hasDivisor n (d+1)) | _ -> false let IsPrime2 n = hasDivisor n 2 |> not let SumOfPrimes max = [|2..max|] |> Array.filter IsPrime1 |> Array.sum let maxVal = 20000 let s = new Stopwatch() s.Start() let valOfSum = SumOfPrimes maxVal s.Stop() Console.WriteLine valOfSum Console.WriteLine("IsPrime1: {0}", s.ElapsedMilliseconds) ////////////////////////////////// s.Reset() s.Start() let SumOfPrimes2 max = [|2..max|] |> Array.filter IsPrime2 |> Array.sum let valOfSum2 = SumOfPrimes2 maxVal s.Stop() Console.WriteLine valOfSum2 Console.WriteLine("IsPrime2: {0}", s.ElapsedMilliseconds) Console.ReadKey() IsPrime1 takes 760 ms while IsPrime2 takes 260ms for the same result. What's going on here and how I can make my code even faster?

  Read the article

• Linux RAID-0 performance doesn't scale up over 1 GB/s

  - by wazoox

  I have trouble getting the max throughput out of my setup. The hardware is as follow : dual Quad-Core AMD Opteron(tm) Processor 2376 16 GB DDR2 ECC RAM dual Adaptec 52245 RAID controllers 48 1 TB SATA drives set up as 2 RAID-6 arrays (256KB stripe) + spares. Software : Plain vanilla 2.6.32.25 kernel, compiled for AMD-64, optimized for NUMA; Debian Lenny userland. benchmarks run : disktest, bonnie++, dd, etc. All give the same results. No discrepancy here. io scheduler used : noop. Yeah, no trick here. Up until now I basically assumed that striping (RAID 0) several physical devices should augment performance roughly linearly. However this is not the case here : each RAID array achieves about 780 MB/s write, sustained, and 1 GB/s read, sustained. writing to both RAID arrays simultaneously with two different processes gives 750 + 750 MB/s, and reading from both gives 1 + 1 GB/s. however when I stripe both arrays together, using either mdadm or lvm, the performance is about 850 MB/s writing and 1.4 GB/s reading. at least 30% less than expected! running two parallel writer or reader processes against the striped arrays doesn't enhance the figures, in fact it degrades performance even further. So what's happening here? Basically I ruled out bus or memory contention, because when I run dd on both drives simultaneously, aggregate write speed actually reach 1.5 GB/s and reading speed tops 2 GB/s. So it's not the PCIe bus. I suppose it's not the RAM. It's not the filesystem, because I get exactly the same numbers benchmarking against the raw device or using XFS. And I also get exactly the same performance using either LVM striping and md striping. What's wrong? What's preventing a process from going up to the max possible throughput? Is Linux striping defective? What other tests could I run?

  Read the article

• mounting ext4 fs with block size of 65536

  - by seaquest

  I am doing some benchmarking on EXT4 performance on Compact Flash media. I have created an ext4 fs with block size of 65536. however I can not mount it on ubuntu-10.10-netbook-i386. (it is already mounting ext4 fs with 4096 bytes of block sizes) According to my readings on ext4 it should allow such big block sized fs. I want to hear your comments. root@ubuntu:~# mkfs.ext4 -b 65536 /dev/sda3 Warning: blocksize 65536 not usable on most systems. mke2fs 1.41.12 (17-May-2010) mkfs.ext4: 65536-byte blocks too big for system (max 4096) Proceed anyway? (y,n) y Warning: 65536-byte blocks too big for system (max 4096), forced to continue Filesystem label= OS type: Linux Block size=65536 (log=6) Fragment size=65536 (log=6) Stride=0 blocks, Stripe width=0 blocks 19968 inodes, 19830 blocks 991 blocks (5.00%) reserved for the super user First data block=0 1 block group 65528 blocks per group, 65528 fragments per group 19968 inodes per group Writing inode tables: done Creating journal (1024 blocks): done Writing superblocks and filesystem accounting information: done This filesystem will be automatically checked every 37 mounts or 180 days, whichever comes first. Use tune2fs -c or -i to override. root@ubuntu:~# tune2fs -l /dev/sda3 tune2fs 1.41.12 (17-May-2010) Filesystem volume name: <none> Last mounted on: <not available> Filesystem UUID: 4cf3f507-e7b4-463c-be11-5b408097099b Filesystem magic number: 0xEF53 Filesystem revision #: 1 (dynamic) Filesystem features: has_journal ext_attr resize_inode dir_index filetype extent flex_bg sparse_super large_file huge_file uninit_bg dir_nlink extra_isize Filesystem flags: signed_directory_hash Default mount options: (none) Filesystem state: clean Errors behavior: Continue Filesystem OS type: Linux Inode count: 19968 Block count: 19830 Reserved block count: 991 Free blocks: 18720 Free inodes: 19957 First block: 0 Block size: 65536 Fragment size: 65536 Blocks per group: 65528 Fragments per group: 65528 Inodes per group: 19968 Inode blocks per group: 78 Flex block group size: 16 Filesystem created: Sat Feb 5 14:39:55 2011 Last mount time: n/a Last write time: Sat Feb 5 14:40:02 2011 Mount count: 0 Maximum mount count: 37 Last checked: Sat Feb 5 14:39:55 2011 Check interval: 15552000 (6 months) Next check after: Thu Aug 4 14:39:55 2011 Lifetime writes: 70 MB Reserved blocks uid: 0 (user root) Reserved blocks gid: 0 (group root) First inode: 11 Inode size: 256 Required extra isize: 28 Desired extra isize: 28 Journal inode: 8 Default directory hash: half_md4 Directory Hash Seed: afb5b570-9d47-4786-bad2-4aacb3b73516 Journal backup: inode blocks root@ubuntu:~# mount -t ext4 /dev/sda3 /mnt/ mount: wrong fs type, bad option, bad superblock on /dev/sda3, missing codepage or helper program, or other error In some cases useful info is found in syslog - try dmesg | tail or so

  Read the article

• BizTalk host throttling &ndash; Singleton pattern and High database size

  - by S.E.R.

  Originally posted on: http://geekswithblogs.net/SERivas/archive/2013/06/30/biztalk-host-throttling-ndash-singleton-pattern-and-high-database-size.aspxI have worked for some days around the singleton pattern (for those unfamiliar with it, read this post by Victor Fehlberg) and have come across a few very interesting posts, among which one dealt with performance issues (here, also by Victor Fehlberg). Simply put: if you have an orchestration which implements the singleton pattern, then performances will continuously decrease as the orchestration receives and consumes messages, and that behavior is more obvious when the orchestration never ends (ie : it keeps looping and never terminates or completes). As I experienced the same kind of problem (actually I was alerted by SCOM, which told me that the host was being throttled because of High database size), I thought it would be a good idea to dig a little bit a see what happens deep inside BizTalk and thus understand the reasons for this behavior. NOTE: in this article, I will focus on this High database size throttling condition. I will try and work on the other conditions in some not too distant future… Test conditions The singleton orchestration For the purpose of this study, I have created the following orchestration, which is a very basic implementation of a singleton that piles up incoming messages, then does something else when a certain timeout has been reached without receiving another message: Throttling settings I have two distinct hosts : one that hosts the receive port (basic FILE port) : Ports_ReceiveHostone that hosts the orchestration : ProcessingHost In order to emphasize the throttling mechanism, I have modified the throttling settings for each of these hosts are as follows (all other parameters are set to the default value): [Throttling thresholds] Message count in database: 500 (default value : 50000) Evolution of performance counters when submitting messages Since we are investigating the High database size throttling condition, here are the performance counter that we should take a look at (all of them are in the BizTalk:Message Agent performance object): Database sizeHigh database sizeMessage delivery throttling stateMessage publishing throttling stateMessage delivery delay (ms)Message publishing delay (ms)Message delivery throttling state durationMessage publishing throttling state duration (If you are not used to Perfmon, I strongly recommend that you start using it right now: it is a wonderful tool that allows you to open the hood and see what is going on inside BizTalk – and other systems) Database size It is quite obvious that we will start by watching the database size and high database size counters, just to see when the first reaches the configured threshold (500) and when the second rings the alarm. NOTE : During this test I submitted 600 messages, one message at a time every 10ms to see the evolution of the counters we have previously selected. It might not show very well on this screenshot, but here is what happened: From 15:46:50 to 15:47:50, the database size for the Ports_ReceiveHost host (blue line) kept growing until it reached a maximum of 504.At 15:47:50, the high database size alert fires At first I was surprised by this result: why is it the database size of the receiving host that keeps growing since it is the processing host that piles up messages? Actually, it makes total sense. This counter measures the size of the database queue that is being filled by the host, not consumed. Therefore, the high database size alert is raised on the host that fills the queue: Ports_ReceiveHost. More information is available on the Public MPWiki page. Now, looking at the Message publishing throttling state for the receiving host (green line), we can see that a throttling condition has been reached at 15:47:50: We can also see that the Message publishing delay(ms) (blue line) has begun growing slowly from this point. All of this explains why performances keep decreasing when a singleton keeps processing new messages: the database size grows and when it has exceeded the Message count in database threshold, the host is throttled and the publishing delay keeps increasing. Digging further So, what happens to the database queue then? Is it flushed some day or does it keep growing and growing indefinitely? The real question being: will the host be throttled forever because of this singleton? To answer this question, I set the Message count in database threshold to 20 (this value is very low in order not to wait for too long, otherwise I certainly would have fallen asleep in front of my screen) and I submitted 30 messages. The test was started at 18:26. At 18:56 (ie : exactly 30min later) the throttling was stopped and the database size was divided by 2. 30 min later again, the database size had dropped to almost zero: I guess I’ll have to find some documentation and do some more testing before I sort this out! My guess is that some maintenance job is at work here, though I cannot tell which one Digging even further If we take a look at the Message delivery throttling state counter for the processing host, we can see that this host was also throttled during the submission of the 600 documents: The value for the counter was 1, meaning that Message delivery incoming rate for the host instance exceeds the Message delivery outgoing rate * the specified Rate overdrive factor (percent) value. We will see this another day… :) A last word Let’s end this article with a warning: DO NOT CHANGE THE THROTTLING SETTINGS LIGHTLY! The temptation can be great to just bypass throttling by setting very high values for each parameter (or zero in some cases, which simply disables throttling). Nevertheless, always keep in mind that this mechanism is here for a very good reason: prevent your BizTalk infrastructure from exploding!! So whatever you do with those settings, do a lot of testing and benchmarking!

  Read the article

• Why lock-free data structures just aren't lock-free enough

  - by Alex.Davies

  Today's post will explore why the current ways to communicate between threads don't scale, and show you a possible way to build scalable parallel programming on top of shared memory. The problem with shared memory Soon, we will have dozens, hundreds and then millions of cores in our computers. It's inevitable, because individual cores just can't get much faster. At some point, that's going to mean that we have to rethink our architecture entirely, as millions of cores can't all access a shared memory space efficiently. But millions of cores are still a long way off, and in the meantime we'll see machines with dozens of cores, struggling with shared memory. Alex's tip: The best way for an application to make use of that increasing parallel power is to use a concurrency model like actors, that deals with synchronisation issues for you. Then, the maintainer of the actors framework can find the most efficient way to coordinate access to shared memory to allow your actors to pass messages to each other efficiently. At the moment, NAct uses the .NET thread pool and a few locks to marshal messages. It works well on dual and quad core machines, but it won't scale to more cores. Every time we use a lock, our core performs an atomic memory operation (eg. CAS) on a cell of memory representing the lock, so it's sure that no other core can possibly have that lock. This is very fast when the lock isn't contended, but we need to notify all the other cores, in case they held the cell of memory in a cache. As the number of cores increases, the total cost of a lock increases linearly. A lot of work has been done on "lock-free" data structures, which avoid locks by using atomic memory operations directly. These give fairly dramatic performance improvements, particularly on systems with a few (2 to 4) cores. The .NET 4 concurrent collections in System.Collections.Concurrent are mostly lock-free. However, lock-free data structures still don't scale indefinitely, because any use of an atomic memory operation still involves every core in the system. A sync-free data structure Some concurrent data structures are possible to write in a completely synchronization-free way, without using any atomic memory operations. One useful example is a single producer, single consumer (SPSC) queue. It's easy to write a sync-free fixed size SPSC queue using a circular buffer*. Slightly trickier is a queue that grows as needed. You can use a linked list to represent the queue, but if you leave the nodes to be garbage collected once you're done with them, the GC will need to involve all the cores in collecting the finished nodes. Instead, I've implemented a proof of concept inspired by this intel article which reuses the nodes by putting them in a second queue to send back to the producer. * In all these cases, you need to use memory barriers correctly, but these are local to a core, so don't have the same scalability problems as atomic memory operations. Performance tests I tried benchmarking my SPSC queue against the .NET ConcurrentQueue, and against a standard Queue protected by locks. In some ways, this isn't a fair comparison, because both of these support multiple producers and multiple consumers, but I'll come to that later. I started on my dual-core laptop, running a simple test that had one thread producing 64 bit integers, and another consuming them, to measure the pure overhead of the queue. So, nothing very interesting here. Both concurrent collections perform better than the lock-based one as expected, but there's not a lot to choose between the ConcurrentQueue and my SPSC queue. I was a little disappointed, but then, the .NET Framework team spent a lot longer optimising it than I did. So I dug out a more powerful machine that Red Gate's DBA tools team had been using for testing. It is a 6 core Intel i7 machine with hyperthreading, adding up to 12 logical cores. Now the results get more interesting. As I increased the number of producer-consumer pairs to 6 (to saturate all 12 logical cores), the locking approach was slow, and got even slower, as you'd expect. What I didn't expect to be so clear was the drop-off in performance of the lock-free ConcurrentQueue. I could see the machine only using about 20% of available CPU cycles when it should have been saturated. My interpretation is that as all the cores used atomic memory operations to safely access the queue, they ended up spending most of the time notifying each other about cache lines that need invalidating. The sync-free approach scaled perfectly, despite still working via shared memory, which after all, should still be a bottleneck. I can't quite believe that the results are so clear, so if you can think of any other effects that might cause them, please comment! Obviously, this benchmark isn't realistic because we're only measuring the overhead of the queue. Any real workload, even on a machine with 12 cores, would dwarf the overhead, and there'd be no point worrying about this effect. But would that be true on a machine with 100 cores? Still to be solved. The trouble is, you can't build many concurrent algorithms using only an SPSC queue to communicate. In particular, I can't see a way to build something as general purpose as actors on top of just SPSC queues. Fundamentally, an actor needs to be able to receive messages from multiple other actors, which seems to need an MPSC queue. I've been thinking about ways to build a sync-free MPSC queue out of multiple SPSC queues and some kind of sign-up mechanism. Hopefully I'll have something to tell you about soon, but leave a comment if you have any ideas.

  Read the article

• In Social Relationship Management, the Spirit is Willing, but Execution is Weak

  - by Mike Stiles

  In our final talk in this series with Aberdeen’s Trip Kucera, we wanted to find out if enterprise organizations are actually doing anything about what they’re learning around the importance of communicating via social and using social listening for a deeper understanding of customers and prospects. We found out that if your brand is lagging behind, you’re not alone. Spotlight: How was Aberdeen able to find out if companies are putting their money where their mouth is when it comes to implementing social across the enterprise? Trip: One way to think about the relative challenges a business has in a given area is to look at the gap between “say” and “do.” The first of those words reveals the brand’s priorities, while the second reveals their ability to execute on those priorities. In Aberdeen’s research, we capture this by asking firms to rank the value of a set of activities from one on the low end to five on the high end. We then ask them to rank their ability to execute those same activities, again on a one to five, not effective to highly effective scale. Spotlight: And once you get their self-assessments, what is it you’re looking for? Trip: There are two things we’re looking for in this analysis. The first is we want to be able to identify the widest gaps between perception of value and execution. This suggests impediments to adoption or simply a high level of challenge, be it technical or otherwise. It may also suggest areas where we can expect future investment and innovation. Spotlight: So the biggest potential pain points surface, places where they know something is critical but also know they aren’t doing much about it. What’s the second thing you look for? Trip: The second thing we want to do is look at specific areas in which high-performing companies, the Leaders, are out-executing the Followers. This points to the business impact of these activities since Leaders are defined by a set of business performance metrics. Put another way, we’re correlating adoption of specific business competencies with performance, looking for what high-performers do differently. Spotlight: Ah ha, that tells us what steps the winners are taking that are making them winners. So what did you find out? Trip: Generally speaking, we see something of a glass curtain when it comes to the social relationship management execution gap. There isn’t a single social media activity in which more than 50% of respondents indicated effectiveness, which would be a 4 or 5 on that 1-5 scale. This despite the fact that 70% of firms indicate that generating positive social media mentions is valuable or very valuable, a 4 or 5 on our 1-5 scale. Spotlight: Well at least they get points for being honest. The verdict they’re giving themselves is that they just aren’t cutting it in these highly critical social development areas. Trip: And the widest gap is around directly engaging with customers and/or prospects on social networks, which 69% of firms rated as valuable but only 34% of companies say they are executing well. Perhaps even more interesting is that these two are interdependent since you’re most likely to generate goodwill on social through happy, engaged customers. This data also suggests that social is largely being used as a broadcast channel rather than for one-to-one engagement. As we’ve discussed previously, social is an inherently personal media. Spotlight: And if they’re still using it as a broadcast channel, that shows they still fail to understand the root of social and see it as just another outlet for their ads and push-messaging. That’s depressing. Trip: A second way to evaluate this data is by using Aberdeen’s performance benchmarking. The story is both a bit different, but consistent in its own way. The first thing we notice is that Leaders are more effective in their execution of several key social relationship management capabilities, namely generating positive mentions and engaging with “influencers” and customers. Based on the fact that Aberdeen uses a broad set of performance metrics to rank the respondents as either “Leaders” (top 35% in weighted performance) or “Followers” (bottom 65% in weighted performance), from website conversion to annual revenue growth, we can then correlated high social effectiveness with company performance. We can also connect the specific social capabilities used by Leaders with effectiveness. We spoke about a few of those key capabilities last time and also discuss them in a new report: Social Powers Activate: Engineering Social Engagement to Win the Hidden Sales Cycle. Spotlight: What all that tells me is there are rewards for making the effort and getting it right. That’s how you become a Leader. Trip: But there’s another part of the story, which is that overall effectiveness, even among Leaders, is muted. There’s just one activity in which more than a majority of Leaders cite high effectiveness, effectiveness being the generation of positive buzz. While 80% of Leaders indicate “directly engaging with customers” through social media channels is valuable, the highest rated activity among Leaders, only 42% say they’re effective. This gap even among Leaders shows the challenges still involved in effective social relationship management. @mikestilesPhoto: stock.xchng

  Read the article

• Siemens AG, Sector Healthcare, Increases Transparency and Improves Customer Loyalty with Web Portal Solution

  - by Kellsey Ruppel

  Siemens AG, Sector Healthcare, Increases Transparency and Improves Customer Loyalty with Web Portal Solution CUSTOMER AND PARTNER INFORMATION Customer Name – Siemens AG, Sector Healthcare Customer Revenue – 73,515 Billion Euro (2011, Siemens AG total) Customer Quote – “The realization of our complex requirements within a very short amount of time was enabled through the competent implementation partner Sapient, who fully used the  very broad scope of standard functionality provided in the Oracle WebCenter Portal, and the management of customer services, who continuously supported the project setup. ” – Joerg Modlmayr, Project Manager, Healthcare Customer Service Portal, Siemens AG The Siemens Healthcare Sector is one of the world's largest suppliers to the healthcare industry and a trendsetter in medical imaging, laboratory diagnostics, medical information technology and hearing aids. Siemens offers its customers products and solutions for the entire range of patient care from a single source – from prevention and early detection to diagnosis, and on to treatment and aftercare. By optimizing clinical workflows for the most common diseases, Siemens also makes healthcare faster, better and more cost-effective. To ensure greater transparency, increased efficiency, higher user acceptance, and additional services, Siemens AG, Sector Healthcare, replaced several existing legacy portal solutions that could not meet the company’s future needs with Oracle WebCenter Portal. Various existing portal solutions that cannot meet future demands will be successively replaced by the new central service portal, which will also allow for the efficient and intuitive implementation of new service concepts.  With Oracle, doctors and hospitals using Siemens medical solutions now have access to a central information portal that provides important information and services at just the push of a button.  Customer Name – Siemens AG, Sector Healthcare Customer URL – www.siemens.com Customer Headquarters – Erlangen, Germany Industry – Industrial Manufacturing Employees – 360,000  Challenges – Replace disparate medical service portals to meet future demands and eliminate an  unnecessarily high level of administrative work caused by heterogeneous installations Ensure portals meet current user demands to improve user-acceptance rates and increase number of total users Enable changes and expansion through standard functionality to eliminate the need for reliance on IT and reduce administrative efforts and associated high costs Ensure efficient and intuitive implementation of new service concepts for all devices and systems Ensure hospitals and clinics to transparently monitor and measure services rendered for the various medical devices and systems  Increase electronic interaction and expand services to achieve a higher level of customer loyalty Solution –  Deployed Oracle WebCenter Portal to ensure greater transparency, and as a result, a higher level of customer loyalty  Provided a centralized platform for doctors and hospitals using Siemens’ medical technology solutions that provides important information and services at the push of a button Reduced significantly the administrative workload by centralizing the solution in the new customer service portal Secured positive feedback from customers involved in the pilot program developed by design experts from Oracle partner Sapient. The interfaces were created with customer needs in mind. The first survey taken shortly after implementation came back with 2.4 points on a scale of 0-3 in the category “customer service portal intuitiveness level” Met all requirements including alignment with the Siemens Style Guide without extensive programming Implemented additional services via the portal such as benchmarking options to ensure the optimal use of the Customer Device Park Provided option for documentation of all services rendered in conjunction with the medical technology systems to ensure that the value of the services are transparent for the decision makers in the hospitals  Saved and stored all machine data from approximately 100,000 remote systems in the central service and information platform Provided the option to register errors online and follow the call status in real-time on the portal Made  available at the push of a button all information on the medical technology devices used in hospitals or clinics—from security checks and maintenance activities to current device statuses Provided PDF format Service Performance Reports that summarize information from periods of time ranging from previous weeks up to one year, meeting medical product law requirements  Why Oracle – Siemens AG favored Oracle for many reasons, however, the company ultimately decided to go with Oracle due to the enormous range of functionality the solutions offered for the healthcare sector.“We are not programmers; we are service providers in the medical technology segment and focus on the contents of the portal. All the functionality necessary for internet-based customer interaction is already standard in Oracle WebCenter Portal, which is a huge plus for us. Having Oracle as our technology partner ensures that the product will continually evolve, providing a strong technology platform for our customer service portal well into the future,” said Joerg Modlmayr project manager, Healthcare Customer Service Portal, Siemens AG. Partner Involvement – Siemens AG selected Oracle Partner Sapient because the company offered a service portfolio that perfectly met Siemens’ requirements and had a wealth of experience implementing Oracle WebCenter Portal. Additionally, Sapient had designers with a very high level of expertise in usability—an aspect that Siemens considered to be of vast importance for the project.  “The Sapient team completely met all our expectations. Our tightly timed project was completed on schedule, and the positive feedback from our users proves that we set the right measures in terms of usability—all thanks to the folks at Sapient,” Modlmayr said.  Partner Name – Sapient GmbH Deutschland Partner URL – www.sapient.com

  Read the article

• Take Two: Comparing JVMs on ARM/Linux

  - by user12608080

  Although the intent of the previous article, entitled Comparing JVMs on ARM/Linux, was to introduce and highlight the availability of the HotSpot server compiler (referred to as c2) for Java SE-Embedded ARM v7,  it seems, based on feedback, that everyone was more interested in the OpenJDK comparisons to Java SE-E.  In fact there were two main concerns: The fact that the previous article compared Java SE-E 7 against OpenJDK 6 might be construed as an unlevel playing field because version 7 is newer and therefore potentially more optimized. That the generic compiler settings chosen to build the OpenJDK implementations did not put those versions in a particularly favorable light. With those considerations in mind, we'll institute the following changes to this version of the benchmarking: In order to help alleviate an additional concern that there is some sort of benchmark bias, we'll use a different suite, called DaCapo.  Funded and supported by many prestigious organizations, DaCapo's aim is to benchmark real world applications.  Further information about DaCapo can be found at http://dacapobench.org. At the suggestion of Xerxes Ranby, who has been a great help through this entire exercise, a newer Linux distribution will be used to assure that the OpenJDK implementations were built with more optimal compiler settings.  The Linux distribution in this instance is Ubuntu 11.10 Oneiric Ocelot. Having experienced difficulties getting Ubuntu 11.10 to run on the original D2Plug ARMv7 platform, for these benchmarks, we'll switch to an embedded system that has a supported Ubuntu 11.10 release.  That platform is the Freescale i.MX53 Quick Start Board.  It has an ARMv7 Coretex-A8 processor running at 1GHz with 1GB RAM. We'll limit comparisons to 4 JVM implementations: Java SE-E 7 Update 2 c1 compiler (default) Java SE-E 6 Update 30 (c1 compiler is the only option) OpenJDK 6 IcedTea6 1.11pre 6b23~pre11-0ubuntu1.11.10.2 CACAO build 1.1.0pre2 OpenJDK 6 IcedTea6 1.11pre 6b23~pre11-0ubuntu1.11.10.2 JamVM build-1.6.0-devel Certain OpenJDK implementations were eliminated from this round of testing for the simple reason that their performance was not competitive.  The Java SE 7u2 c2 compiler was also removed because although quite respectable, it did not perform as well as the c1 compilers.  Recall that c2 works optimally in long-lived situations.  Many of these benchmarks completed in a relatively short period of time.  To get a feel for where c2 shines, take a look at the first chart in this blog. The first chart that follows includes performance of all benchmark runs on all platforms.  Later on we'll look more at individual tests.  In all runs, smaller means faster.  The DaCapo aficionado may notice that only 10 of the 14 DaCapo tests for this version were executed.  The reason for this is that these 10 tests represent the only ones successfully completed by all 4 JVMs.  Only the Java SE-E 6u30 could successfully run all of the tests.  Both OpenJDK instances not only failed to complete certain tests, but also experienced VM aborts too. One of the first observations that can be made between Java SE-E 6 and 7 is that, for all intents and purposes, they are on par with regards to performance.  While it is a fact that successive Java SE releases add additional optimizations, it is also true that Java SE 7 introduces additional complexity to the Java platform thus balancing out any potential performance gains at this point.  We are still early into Java SE 7.  We would expect further performance enhancements for Java SE-E 7 in future updates. In comparing Java SE-E to OpenJDK performance, among both OpenJDK VMs, Cacao results are respectable in 4 of the 10 tests.  The charts that follow show the individual results of those four tests.  Both Java SE-E versions do win every test and outperform Cacao in the range of 9% to 55%. For the remaining 6 tests, Java SE-E significantly outperforms Cacao in the range of 114% to 311% So it looks like OpenJDK results are mixed for this round of benchmarks.  In some cases, performance looks to have improved.  But in a majority of instances, OpenJDK still lags behind Java SE-Embedded considerably. Time to put on my asbestos suit.  Let the flames begin...

  Read the article

• Performance triage

  - by Dave

  Folks often ask me how to approach a suspected performance issue. My personal strategy is informed by the fact that I work on concurrency issues. (When you have a hammer everything looks like a nail, but I'll try to keep this general). A good starting point is to ask yourself if the observed performance matches your expectations. Expectations might be derived from known system performance limits, prototypes, and other software or environments that are comparable to your particular system-under-test. Some simple comparisons and microbenchmarks can be useful at this stage. It's also useful to write some very simple programs to validate some of the reported or expected system limits. Can that disk controller really tolerate and sustain 500 reads per second? To reduce the number of confounding factors it's better to try to answer that question with a very simple targeted program. And finally, nothing beats having familiarity with the technologies that underlying your particular layer. On the topic of confounding factors, as our technology stacks become deeper and less transparent, we often find our own technology working against us in some unexpected way to choke performance rather than simply running into some fundamental system limit. A good example is the warm-up time needed by just-in-time compilers in Java Virtual Machines. I won't delve too far into that particular hole except to say that it's rare to find good benchmarks and methodology for java code. Another example is power management on x86. Power management is great, but it can take a while for the CPUs to throttle up from low(er) frequencies to full throttle. And while I love "turbo" mode, it makes benchmarking applications with multiple threads a chore as you have to remember to turn it off and then back on otherwise short single-threaded runs may look abnormally fast compared to runs with higher thread counts. In general for performance characterization I disable turbo mode and fix the power governor at "performance" state. Another source of complexity is the scheduler, which I've discussed in prior blog entries. Lets say I have a running application and I want to better understand its behavior and performance. We'll presume it's warmed up, is under load, and is an execution mode representative of what we think the norm would be. It should be in steady-state, if a steady-state mode even exists. On Solaris the very first thing I'll do is take a set of "pstack" samples. Pstack briefly stops the process and walks each of the stacks, reporting symbolic information (if available) for each frame. For Java, pstack has been augmented to understand java frames, and even report inlining. A few pstack samples can provide powerful insight into what's actually going on inside the program. You'll be able to see calling patterns, which threads are blocked on what system calls or synchronization constructs, memory allocation, etc. If your code is CPU-bound then you'll get a good sense where the cycles are being spent. (I should caution that normal C/C++ inlining can diffuse an otherwise "hot" method into other methods. This is a rare instance where pstack sampling might not immediately point to the key problem). At this point you'll need to reconcile what you're seeing with pstack and your mental model of what you think the program should be doing. They're often rather different. And generally if there's a key performance issue, you'll spot it with a moderate number of samples. I'll also use OS-level observability tools to lock for the existence of bottlenecks where threads contend for locks; other situations where threads are blocked; and the distribution of threads over the system. On Solaris some good tools are mpstat and too a lesser degree, vmstat. Try running "mpstat -a 5" in one window while the application program runs concurrently. One key measure is the voluntary context switch rate "vctx" or "csw" which reflects threads descheduling themselves. It's also good to look at the user; system; and idle CPU percentages. This can give a broad but useful understanding if your threads are mostly parked or mostly running. For instance if your program makes heavy use of malloc/free, then it might be the case you're contending on the central malloc lock in the default allocator. In that case you'd see malloc calling lock in the stack traces, observe a high csw/vctx rate as threads block for the malloc lock, and your "usr" time would be less than expected. Solaris dtrace is a wonderful and invaluable performance tool as well, but in a sense you have to frame and articulate a meaningful and specific question to get a useful answer, so I tend not to use it for first-order screening of problems. It's also most effective for OS and software-level performance issues as opposed to HW-level issues. For that reason I recommend mpstat & pstack as my the 1st step in performance triage. If some other OS-level issue is evident then it's good to switch to dtrace to drill more deeply into the problem. Only after I've ruled out OS-level issues do I switch to using hardware performance counters to look for architectural impediments.

  Read the article

• Why do many software projects fail today?

  - by TomTom

  As long as there are software projects, the world is wondering why they fail so often. I would like to know if there is a list or something equivalent which shows how many software projects fail today. Would be nice if there would be a comparison over the last 20 - 30 years. You can also add your top reason why a software project fails. Mine is "Requirements are poor or not even existing." which includes also "No (real) customer / user involved". EDIT: It is nearly impossible to clearly define the term "fail". Let's say that fail means: The project was more than 10% over budget and time. In my opinion the 10% + / - is a good range for an offer / tender. EDIT: Until now (Feb 11) it seems that most posters agree that a fail of the project is basically a failure of the project management (whatever fail means). But IMHO it comes out, that most developers are not happy with this situation. Perhaps because not the manager get penalized when a project was not successful, but the lazy, incompetent developer teams? When I read the posts I can also hear-out that there is a big "gap" between the developer side and the managment side. The expectations (perhaps also the requirements) seem to be so different, that a project cannot be successful in the end (over time / budget; users are not happy; not all first-prio features implemented; too many bugs because developers were forced to implement in too short timeframes ...) I',m asking myself: How can we improve it? Or do we have the possibility to improve it? Everybody seems to be unsatisfied with the way it goes now. Can we close the gap between these two worlds? Should we (the developers) go on strike and fight for "high quality reqiurements" and "realistic / iteration based time shedules"? EDIT: Ralph Westphal and Stefan Lieser have founded a new "community" called: Clean-Code-Developer. The aim of the group is to bring more professionalism into software engineering. Independently if a developer has a degree or tons of years of experience you can be part of this movement. Clean Code Developers live principles like SOLID every day. A professional developer is the biggest reviewer of his own work. And he has an internal value system which helps him to improve and become better. Check it out on: Clean Code Developer EDIT: Our company is doing at the moment a thing called "Application Development and Maintenance Benchmarking". This is a service offered by IBM to get a feedback from someone external on your software engineering process quality etc. When we get the results, I will tell you more about it.

  Read the article

• Is this slow WPF TextBlock performance expected?

  - by Ben Schoepke

  Hi, I am doing some benchmarking to determine if I can use WPF for a new product. However, early performance results are disappointing. I made a quick app that uses data binding to display a bunch of random text inside of a list box every 100 ms and it was eating up ~15% CPU. So I made another quick app that skipped the data binding/data template scheme and does nothing but update 10 TextBlocks that are inside of a ListBox every 100 ms (the actual product wouldn't require 100 ms updates, more like 500 ms max, but this is a stress test). I'm still seeing ~10-15% CPU usage. Why is this so high? Is it because of all the garbage strings? Here's the XAML: <Grid> <ListBox x:Name="numericsListBox"> <ListBox.Resources> <Style TargetType="TextBlock"> <Setter Property="FontSize" Value="48"/> <Setter Property="Width" Value="300"/> </Style> </ListBox.Resources> <TextBlock/> <TextBlock/> <TextBlock/> <TextBlock/> <TextBlock/> <TextBlock/> <TextBlock/> <TextBlock/> <TextBlock/> <TextBlock/> </ListBox> </Grid> Here's the code behind: public partial class Window1 : Window { private int _count = 0; public Window1() { InitializeComponent(); } private void OnLoad(object sender, RoutedEventArgs e) { var t = new DispatcherTimer(TimeSpan.FromSeconds(0.1), DispatcherPriority.Normal, UpdateNumerics, Dispatcher); t.Start(); } private void UpdateNumerics(object sender, EventArgs e) { ++_count; foreach (object textBlock in numericsListBox.Items) { var t = textBlock as TextBlock; if (t != null) t.Text = _count.ToString(); } } } Any ideas for a better way to quickly render text? My computer: XP SP3, 2.26 GHz Core 2 Duo, 4 GB RAM, Intel 4500 HD integrated graphics. And that is an order of magnitude beefier than the hardware I'd need to develop for in the real product.

  Read the article

• mounting ext4 fs with block size of 65536

  - by seaquest

  I am doing some benchmarking on EXT4 performance on Compact Flash media. I have created an ext4 fs with block size of 65536. however I can not mount it on ubuntu-10.10-netbook-i386. (it is already mounting ext4 fs with 4096 bytes of block sizes) According to my readings on ext4 it should allow such big block sized fs. I want to hear your comments. root@ubuntu:~# mkfs.ext4 -b 65536 /dev/sda3 Warning: blocksize 65536 not usable on most systems. mke2fs 1.41.12 (17-May-2010) mkfs.ext4: 65536-byte blocks too big for system (max 4096) Proceed anyway? (y,n) y Warning: 65536-byte blocks too big for system (max 4096), forced to continue Filesystem label= OS type: Linux Block size=65536 (log=6) Fragment size=65536 (log=6) Stride=0 blocks, Stripe width=0 blocks 19968 inodes, 19830 blocks 991 blocks (5.00%) reserved for the super user First data block=0 1 block group 65528 blocks per group, 65528 fragments per group 19968 inodes per group Writing inode tables: done Creating journal (1024 blocks): done Writing superblocks and filesystem accounting information: done This filesystem will be automatically checked every 37 mounts or 180 days, whichever comes first. Use tune2fs -c or -i to override. root@ubuntu:~# tune2fs -l /dev/sda3 tune2fs 1.41.12 (17-May-2010) Filesystem volume name: <none> Last mounted on: <not available> Filesystem UUID: 4cf3f507-e7b4-463c-be11-5b408097099b Filesystem magic number: 0xEF53 Filesystem revision #: 1 (dynamic) Filesystem features: has_journal ext_attr resize_inode dir_index filetype extent flex_bg sparse_super large_file huge_file uninit_bg dir_nlink extra_isize Filesystem flags: signed_directory_hash Default mount options: (none) Filesystem state: clean Errors behavior: Continue Filesystem OS type: Linux Inode count: 19968 Block count: 19830 Reserved block count: 991 Free blocks: 18720 Free inodes: 19957 First block: 0 Block size: 65536 Fragment size: 65536 Blocks per group: 65528 Fragments per group: 65528 Inodes per group: 19968 Inode blocks per group: 78 Flex block group size: 16 Filesystem created: Sat Feb 5 14:39:55 2011 Last mount time: n/a Last write time: Sat Feb 5 14:40:02 2011 Mount count: 0 Maximum mount count: 37 Last checked: Sat Feb 5 14:39:55 2011 Check interval: 15552000 (6 months) Next check after: Thu Aug 4 14:39:55 2011 Lifetime writes: 70 MB Reserved blocks uid: 0 (user root) Reserved blocks gid: 0 (group root) First inode: 11 Inode size: 256 Required extra isize: 28 Desired extra isize: 28 Journal inode: 8 Default directory hash: half_md4 Directory Hash Seed: afb5b570-9d47-4786-bad2-4aacb3b73516 Journal backup: inode blocks root@ubuntu:~# mount -t ext4 /dev/sda3 /mnt/ mount: wrong fs type, bad option, bad superblock on /dev/sda3, missing codepage or helper program, or other error In some cases useful info is found in syslog - try dmesg | tail or so

  Read the article

• Benchmark of Java Try/Catch Block

  - by hectorg87

  I know that going into a catch block has some significance cost when executing a program, however, I was wondering if entering a try{} block also had any impact so I started looking for an answer in google with many opinions, but no benchmarking at all. Some answers I found were: Java try/catch performance, is it recommended to keep what is inside the try clause to a minimum? Try Catch Performance Java Java try catch blocks However they didn't answer my question with facts, so I decided to try it for myself. Here's what I did. I have a csv file with this format: host;ip;number;date;status;email;uid;name;lastname;promo_code; where everything after status is optional and will not even have the corresponding ; , so when parsing a validation has to be done to see if the value is there, here's where the try/catch issue came to my mind. The current code that in inherited in my company does this: StringTokenizer st=new StringTokenizer(line,";"); String host = st.nextToken(); String ip = st.nextToken(); String number = st.nextToken(); String date = st.nextToken(); String status = st.nextToken(); String email = ""; try{ email = st.nextToken(); }catch(NoSuchElementException e){ email = ""; } and it repeats what it's done for email with uid, name, lastname and promo_code. and I changed everything to: if(st.hasMoreTokens()){ email = st.nextToken(); } and in fact it performs faster. When parsing a file that doesn't have the optional columns. Here are the average times: --- Trying:122 milliseconds --- Checking:33 milliseconds however, here's what confused me and the reason I'm asking: When running the example with values for the optional columns in all 8000 lines of the CSV, the if() version still performs better than the try/catch version, so my question is Does really the try block does not have any performance impact on my code? The average times for this example are: --- Trying:105 milliseconds --- Checking:43 milliseconds Can somebody explain what's going on here? Thanks a lot

  Read the article

• Memcached Lagging

  - by Brad Dwyer

  Let me preface this by saying that this is a followup question to this topic. That was "solved" by switching from Solaris (SmartOS) to Ubuntu for the memcached server. Now we've multiplied load by about 5x and are running into problems again. We are running a site that is doing about 1000 requests/minute, each request hits Memcached with approximately 3 reads and 1 write. So load is approximately 65 requests per second. Total data in the cache is about 37M, and each key contains a very small amount of data (a JSON-encoded array of integers amounting to less than 1K). We have setup a benchmarking script on these pages and fed the data into StatsD for logging. The problem is that there are spikes where Memcached takes a very long time to respond. These do not appear to correlate with spikes in traffic. What could be causing these spikes? Why would memcached take over a second to reply? We just booted up a second server to put in the pool and it didn't make any noticeable difference in the frequency or severity of the spikes. This is the output of getStats() on the servers: Array ( [-----------] => Array ( [pid] => 1364 [uptime] => 3715684 [threads] => 4 [time] => 1336596719 [pointer_size] => 64 [rusage_user_seconds] => 7924 [rusage_user_microseconds] => 170000 [rusage_system_seconds] => 187214 [rusage_system_microseconds] => 190000 [curr_items] => 12578 [total_items] => 53516300 [limit_maxbytes] => 943718400 [curr_connections] => 14 [total_connections] => 72550117 [connection_structures] => 165 [bytes] => 2616068 [cmd_get] => 450388258 [cmd_set] => 53493365 [get_hits] => 450388258 [get_misses] => 2244297 [evictions] => 0 [bytes_read] => 2138744916 [bytes_written] => 745275216 [version] => 1.4.2 ) [-----------:11211] => Array ( [pid] => 8099 [uptime] => 4687 [threads] => 4 [time] => 1336596719 [pointer_size] => 64 [rusage_user_seconds] => 7 [rusage_user_microseconds] => 170000 [rusage_system_seconds] => 290 [rusage_system_microseconds] => 990000 [curr_items] => 2384 [total_items] => 225964 [limit_maxbytes] => 943718400 [curr_connections] => 7 [total_connections] => 588097 [connection_structures] => 91 [bytes] => 562641 [cmd_get] => 1012562 [cmd_set] => 225778 [get_hits] => 1012562 [get_misses] => 125161 [evictions] => 0 [bytes_read] => 91270698 [bytes_written] => 350071516 [version] => 1.4.2 ) ) Edit: Here is the result of a set and retrieve of 10,000 values. Normal: Stored 10000 values in 5.6118 seconds. Average: 0.0006 High: 0.1958 Low: 0.0003 Fetched 10000 values in 5.1215 seconds. Average: 0.0005 High: 0.0141 Low: 0.0003 When Spiking: Stored 10000 values in 16.5074 seconds. Average: 0.0017 High: 0.9288 Low: 0.0003 Fetched 10000 values in 19.8771 seconds. Average: 0.0020 High: 0.9478 Low: 0.0003

  Read the article

• Is Apache 2.2.22 able to sustain 1.000 simultaneous connected clients?

  - by Fnux

  For an article in a news paper, I'm benchmarking 5 different web servers (Apache2, Cherokee, Lighttpd, Monkey and Nginx). The tests made consist of measuring the execution times as well as different parameters such as the number of request served per second, the amount of RAM, the CPU used, during a growing load of simultaneous clients (from 1 to 1.000 with a step of 10) each client sending 1.000.000 requets of a small fixed file, then of a medium fixed file, then a small dynamic content (hello.php) and finally a complex dynamic content (the computation of the reimbursment of a loan). All the web servers are able to sustain such a load (up to 1.000 clients) but Apache2 which always stops to respond when the test reach 450 to 500 simultaneous clients. My configuration is : CPU: AMD FX 8150 8 cores @ 4.2 GHz RAM: 32 Gb. SSD: 2 x Crucial 240 Gb SATA6 OS: Ubuntu 12.04.3 64 bit WS: Apache 2.2.22 My Apache2 configuration is as follows: /etc/apache2/apache2.conf LockFile ${APACHE_LOCK_DIR}/accept.lock PidFile ${APACHE_PID_FILE} Timeout 30 KeepAlive On MaxKeepAliveRequests 1000000 KeepAliveTimeout 2 ServerName "fnux.net" <IfModule mpm_prefork_module> StartServers 16 MinSpareServers 16 MaxSpareServers 16 ServerLimit 2048 MaxClients 1024 MaxRequestsPerChild 0 </IfModule> User ${APACHE_RUN_USER} Group ${APACHE_RUN_GROUP} AccessFileName .htaccess <Files ~ "^\.ht"> Order allow,deny Deny from all Satisfy all </Files> DefaultType None HostnameLookups Off ErrorLog ${APACHE_LOG_DIR}/error.log LogLevel emerg Include mods-enabled/*.load Include mods-enabled/*.conf Include httpd.conf Include ports.conf LogFormat "%v:%p %h %l %u %t \"%r\" %>s %O \"%{Referer}i\" \"%{User-Agent}i\"" vhost_combined LogFormat "%h %l %u %t \"%r\" %>s %O \"%{Referer}i\" \"%{User-Agent}i\"" combined LogFormat "%h %l %u %t \"%r\" %>s %O" common LogFormat "%{Referer}i -> %U" referer LogFormat "%{User-agent}i" agent Include conf.d/ Include sites-enabled/ /etc/apache2/ports.conf NameVirtualHost *:8180 Listen 8180 <IfModule mod_ssl.c> Listen 443 </IfModule> <IfModule mod_gnutls.c> Listen 443 </IfModule> /etc/apache2/mods-available <IfModule mod_fastcgi.c> AddHandler php5-fcgi .php Action php5-fcgi /cgi-bin/php5.external <Location "/cgi-bin/php5.external"> Order Deny,Allow Deny from All Allow from env=REDIRECT_STATUS </Location> </IfModule> /etc/apache2/sites-available/default <VirtualHost *:8180> ServerAdmin webmaster@localhost DocumentRoot /var/www/apache2 <Directory /> Options FollowSymLinks AllowOverride None </Directory> <Directory /var/www/> Options Indexes FollowSymLinks MultiViews AllowOverride None Order allow,deny allow from all </Directory> ScriptAlias /cgi-bin/ /usr/lib/cgi-bin/ <Directory "/usr/lib/cgi-bin"> AllowOverride None Options +ExecCGI -MultiViews +SymLinksIfOwnerMatch Order allow,deny Allow from all </Directory> ErrorLog ${APACHE_LOG_DIR}/error.log LogLevel emerg ##### CustomLog ${APACHE_LOG_DIR}/access.log combined Alias /doc/ "/usr/share/doc/" <Directory "/usr/share/doc/"> Options Indexes MultiViews FollowSymLinks AllowOverride None Order deny,allow Deny from all Allow from 127.0.0.0/255.0.0.0 ::1/128 </Directory> <IfModule mod_fastcgi.c> AddHandler php5-fcgi .php Action php5-fcgi /php5-fcgi Alias /php5-fcgi /usr/lib/cgi-bin/php5-fcgi FastCgiExternalServer /usr/lib/cgi-bin/php5-fcgi -host 127.0.0.1:9000 -pass-header Authorization </IfModule> </VirtualHost> /etc/security/limits.conf * soft nofile 1000000 * hard nofile 1000000 So, I would trully appreciate your advice to setup Apache2 to make it able to sustain 1.000 simultaneous clients, if this is even possible. TIA for your help. Cheers.

  Read the article

• Does this prove a network bandwidth bottleneck?

  - by Yuji Tomita

  I've incorrectly assumed that my internal AB testing means my server can handle 1k concurrency @3k hits per second. My theory at at the moment is that the network is the bottleneck. The server can't send enough data fast enough. External testing from blitz.io at 1k concurrency shows my hits/s capping off at 180, with pages taking longer and longer to respond as the server is only able to return 180 per second. I've served a blank file from nginx and benched it: it scales 1:1 with concurrency. Now to rule out IO / memcached bottlenecks (nginx normally pulls from memcached), I serve up a static version of the cached page from the filesystem. The results are very similar to my original test; I'm capped at around 180 RPS. Splitting the HTML page in half gives me double the RPS, so it's definitely limited by the size of the page. If I internally ApacheBench from the local server, I get consistent results of around 4k RPS on both the Full Page and the Half Page, at high transfer rates. Transfer rate: 62586.14 [Kbytes/sec] received If I AB from an external server, I get around 180RPS - same as the blitz.io results. How do I know it's not intentional throttling? If I benchmark from multiple external servers, all results become poor which leads me to believe the problem is in MY servers outbound traffic, not a download speed issue with my benchmarking servers / blitz.io. So I'm back to my conclusion that my server can't send data fast enough. Am I right? Are there other ways to interpret this data? Is the solution/optimization to set up multiple servers + load balancing that can each serve 180 hits per second? I'm quite new to server optimization, so I'd appreciate any confirmation interpreting this data. Outbound traffic Here's more information about the outbound bandwidth: The network graph shows a maximum output of 16 Mb/s: 16 megabits per second. Doesn't sound like much at all. Due to a suggestion about throttling, I looked into this and found that linode has a 50mbps cap (which I'm not even close to hitting, apparently). I had it raised to 100mbps. Since linode caps my traffic, and I'm not even hitting it, does this mean that my server should indeed be capable of outputting up to 100mbps but is limited by some other internal bottleneck? I just don't understand how networks at this large of a scale work; can they literally send data as fast as they can read from the HDD? Is the network pipe that big? In conclusion 1: Based on the above, I'm thinking I can definitely raise my 180RPS by adding an nginx load balancer on top of a multi nginx server setup at exactly 180RPS per server behind the LB. 2: If linode has a 50/100mbit limit that I'm not hitting at all, there must be something I can do to hit that limit with my single server setup. If I can read / transmit data fast enough locally, and linode even bothers to have a 50mbit/100mbit cap, there must be an internal bottleneck that's not allowing me to hit those caps that I'm not sure how to detect. Correct? I realize the question is huge and vague now, but I'm not sure how to condense it. Any input is appreciated on any conclusion I've made.

  Read the article

• Performance of Cluster Shared Volume file copy from SAN

  - by Sequenzia

  I am hoping someone can help me out with a strange issue. We are running a Microsoft Failover Cluster with Server 2008 R2 and an Equallogic PS4000 SAN. Our main configuration has 2 Dell Poweredge T710 Servers in the cluster. We have CSV and Quorm setup. The servers each have 10 Broadcom 1Gb NICs. Right now 4 of the NICS are on the iSCSI network for accessing the SAN. They use MPIO and the Dell HIT pack. We have 5 VMs running on each node and everything runs smooth. No noticeable performance issues or anything. From the SAN I can see the 4 iSCSI connections from each server to each volume (CSV and Quorm). Again, it seems to perform great. The problem I am running into is with backups. I have tried a few backup programs like backupchain and Veeam. The problem is both of them are very very slow to backup the VMs. For instance I have a 500GB (fixed disc) VHD that’s running on the cluster. It takes over 18 hours to backup that VHD and that’s with compression and depuping turned off which is supposed to be the fasted. We also have a separate server that is just for backups. It has a lot of directed attached storage. As part of the troubleshooting I decided to bring that server into the cluster as a node. It now has access to the CSV and can read from C:\clusterstorage\volume1 which is where our VHDs live. This backup server only has 2 NICs. 1 NIC is going to the iSCSI network and the other is just on the main network. It has Intel NICS in it without any sort of MPIO or teaming. So with the 3rd server now in the cluster I started doing some benchmarking. I have a test VHD that’s about 7GBs that’s stored in the CSV. I have tested file copying that VHD from all 3 servers to directed attached storage in the respective server. The 2 Dell servers that are the main nodes in the cluster (they house the VMs) are reading that file at about 20Mbs/Sec. Which at that rate is way to slow for the backups. The other server which only has 1 NIC to the SAN is reading at around 100Mbs/Sec. I spent a few hours on the phone with Dell today about this . We went through all kind of tests and he was pretty dumb founded. He really has no idea why that server with only 1 NIC is reading about 5 times as fast as the servers with 4 NICS and MPIO. We looked at the network utilization of the NICs while the file copy was going on. The servers with the 4 NICs had a small increase of activity during the file copy but they only went up to around 8-10% on all 4 NICs. The other server with the 1 NIC jumped up to over 80% during the file copy. I plan on doing some more testing after hours and calling Dell back tomorrow but I really am confused (and so is Dell’s support rep) why I cannot get faster file copy access to the CSV on those servers. Anyone have any input on this? Any feedback would be greatly appreciated. Thanks in advance.

  Read the article

• Hardware recommendations / parts list for a modern, quiet ZFS NAS box - 2011-Feb edition [closed]

  - by dandv

  I want to build some really reliable storage for my data, and it seems that ZFS is the only filesystem at the moment that does live checksumming. That rules out DroboPro, so I'm looking to building a quiet ZFS NAS that would start with 4 2TB or larger hard drives. I'd like this system to be very reliable and relatively future-proof for 2-3 years, so I'm willing to invest some $$$ and buy higher end components. I did see questions here and on other forums about low-cost servers, but I'm not looking for those. I'd be super happy to go for an off-the-shelf solution, but I haven't found one that's quiet. I started doing the research (summarized on my wiki), but I realized that it just gets too complicated for what I know as a software dude, and I'm entering the analysis paralysis area. At this point, I'm basically looking for a parts list for a configuration that will work (and is modern), and I know there are folks around here who are way more competent than me. I've built computers and am comfortable assembling one and messing with *nix; I can follow guides; I just want to end the decision process for the hardware and software configuration. What I've researched so far (not that these are very modern components): Case: I think I've settled on the Antec Twelve Hundred case because it cools well, is quiet, and simply has 12 bays that allow elastic mounting. The SilverStone Raven is its counter-candidate, but I find its construction quite odd. For the PSU, I'm torn between Antec CP-850 and Nexus RX-8500, but I did this research more than a year ago. The Nexus has a very uniform power profile, and I'd rather not have the Antec spin up and down based on load. On the other hand, I'm not sure how often my file server will draw more than 400W under use. For the hard drives, I've read that WD Black drives are actually WD RE3 with a software setting changed. I'd also like to buy different drive types, not just 4 WDs. Recommendations? Right now I have a 2TB Hitachi Deskstar 7K300. For the motherboard, CPU and RAM I have no idea, other than the RAM must be ECC. I already asked a question here about ECC RAM, but I was misguided and was looking for a motherboard that would support USB 3.0 as well. I've learned to go with eSATA, or worry about USB later. Then there's the (liquid) cooling, Wi-Fi card, and FreeBSD vs. OpenSolaris Express. Lastly, I'm wondering if I can make this PC into a media server by adding a Blu-ray drive and a good sound card. But support for Blu-Ray is spotty on Linux, and I don't know if Windows 7 on VirtualBox would get sufficient hardware access to output HDMI or SPDIFF signals. (Running OpenSolaris virtualized is not an option because of the reliability risk.) Then there are HDCP concerns. Suggestions on that would be appreciated as well, but I don't want us to get sidetracked. A specific shopping list on the core components would be great, so I can start ordering, and in the meantime educate myself with regards to the other issues. Finally, I think this could become a great FAQ for those technically inclined to build their own ZFS server, but confused by the dizzying array of options out there, and I promise to compile the results and share my experience building and benchmarking the server.

  Read the article

• Solaris 11 Launch Blog Carnival Roundup

  - by constant

  Solaris 11 is here! And together with the official launch activities, a lot of Oracle and non-Oracle bloggers contributed helpful and informative blog articles to help your datacenter go to eleven. Here are some notable blog postings, sorted by category for your Solaris 11 blog-reading pleasure: Getting Started/Overview A lot of people speculated that the official launch of Solaris 11 would be on 11/11 (whatever way you want to turn it), but it actually happened two days earlier. Larry Wake himself offers 11 Reasons Why Oracle Solaris 11 11/11 Isn't Being Released on 11/11/11. Then, Larry goes on with a summary: Oracle Solaris 11: The First Cloud OS gives you a short and sweet rundown of what the major new features of Solaris 11 are. Jeff Victor has his own list of What's New in Oracle Solaris 11. A popular Solaris 11 meme is to write a blog post about 11 favourite features: Jim Laurent's 11 Reasons to Love Solaris 11, Darren Moffat's 11 Favourite Solaris 11 Features, Mike Gerdt's 11 of My Favourite Things! are just three examples of "11 Favourite Things..." type blog posts, I'm sure many more will follow... More official overview content for Solaris 11 is available from the Oracle Tech Network Solaris 11 Portal. Also, check out Rick Ramsey's blog post Solaris 11 Resources for System Administrators on the OTN Blog and his secret 5 Commands That Make Solaris Administration Easier post from the OTN Garage. (Automatic) Installation and the Image Packaging System (IPS) The brand new Image Packaging System (IPS) and the Automatic Installer (IPS), together with numerous other install/packaging/boot/patching features are among the most significant improvements in Solaris 11. But before installing, you may wonder whether Solaris 11 will support your particular set of hardware devices. Again, the OTN Garage comes to the rescue with Rick Ramsey's post How to Find Out Which Devices Are Supported By Solaris 11. Included is a useful guide to all the first steps to get your Solaris 11 system up and running. Tim Foster had a whole handful of blog posts lined up for the launch, teaching you everything you need to know about IPS but didn't dare to ask: The IPS System Repository, IPS Self-assembly - Part 1: Overlays and Part 2: Multiple Packages Delivering Configuration. Watch out for more IPS posts from Tim! If installing packages or upgrading your system from the net makes you uneasy, then you're not alone: Jim Laurent will tech you how Building a Solaris 11 Repository Without Network Connection will make your life easier. Many of you have already peeked into the future by installing Solaris 11 Express. If you're now wondering whether you can upgrade or whether a fresh install is necessary, then check out Alan Hargreaves's post Upgrading Solaris 11 Express b151a with support to Solaris 11. The trick is in upgrading your pkg(1M) first. Networking One of the first things to do after installing Solaris 11 (or any operating system for that matter), is to set it up for networking. Solaris 11 comes with the brand new "Network Auto-Magic" feature which can figure out everything by itself. For those cases where you want to exercise a little more control, Solaris 11 left a few people scratching their heads. Fortunately, Tschokko wrote up this cool blog post: Solaris 11 manual IPv4 & IPv6 configuration right after the launch ceremony. Thanks, Tschokko! And Milek points out a long awaited networking feature in Solaris 11 called Solaris 11 - hostmodel, which I know for a fact that many customers have looked forward to: How to "bind" a Solaris 11 system to a specific gateway for specific IP address it is using. Steffen Weiberle teaches us how to tune the Solaris 11 networking stack the proper way: ipadm(1M). No more fiddling with ndd(1M)! Check out his tutorial on Solaris 11 Network Tunables. And if you want to get even deeper into the networking stack, there's nothing better than DTrace. Alan Maguire teaches you in: DTracing TCP Congestion Control how to probe deeply into the Solaris 11 TCP/IP stack, the TCP congestion control part in particular. Don't miss his other DTrace and TCP related blog posts! DTrace And there we are: DTrace, the king of all observability tools. Long time DTrace veteran and co-author of The DTrace book*, Brendan Gregg blogged about Solaris 11 DTrace syscall provider changes. BTW, after you install Solaris 11, check out the DTrace toolkit which is installed by default in /usr/dtrace/DTT. It is chock full of handy DTrace scripts, many of which contributed by Brendan himself! Security Another big theme in Solaris 11, and one that is crucial for the success of any operating system in the Cloud is Security. Here are some notable posts in this category: Darren Moffat starts by showing us how to completely get rid of root: Completely Disabling Root Logins on Solaris 11. With no root user, there's one major entry point less to worry about. But that's only the start. In Immutable Zones on Encrypted ZFS, Darren shows us how to double the security of your services: First by locking them into the new Immutable Zones feature, then by encrypting their data using the new ZFS encryption feature. And if you're still missing sudo from your Linux days, Darren again has a solution: Password (PAM) caching for Solaris su - "a la sudo". If you're wondering how much compute power all this encryption will cost you, you're in luck: The Solaris X86 AESNI OpenSSL Engine will make sure you'll use your Intel's embedded crypto support to its fullest. And if you own a brand new SPARC T4 machine you're even luckier: It comes with its own SPARC T4 OpenSSL Engine. Dan Anderson's posts show how there really is now excuse not to encrypt any more... Developers Solaris 11 has a lot to offer to developers as well. Ali Bahrami has a series of blog posts that cover diverse developer topics: elffile: ELF Specific File Identification Utility, Using Stub Objects and The Stub Proto: Not Just For Stub Objects Anymore to name a few. BTW, if you're a developer and want to shape the future of Solaris 11, then Vijay Tatkar has a hint for you: Oracle (Sun Systems Group) is hiring! Desktop and Graphics Yes, Solaris 11 is a 100% server OS, but it can also offer a decent desktop environment, especially if you are a developer. Alan Coopersmith starts by discussing S11 X11: ye olde window system in today's new operating system, then Calum Benson shows us around What's new on the Solaris 11 Desktop. Even accessibility is a first-class citizen in the Solaris 11 user interface. Peter Korn celebrates: Accessible Oracle Solaris 11 - released! Performance Gone are the days of "Slowaris", when Solaris was among the few OSes that "did the right thing" while others cut corners just to win benchmarks. Today, Solaris continues doing the right thing, and it delivers the right performance at the same time. Need proof? Check out Brian's BestPerf blog with continuous updates from the benchmarking lab, including Recent Benchmarks Using Oracle Solaris 11! Send Me More Solaris 11 Launch Articles! These are just a few of the more interesting blog articles that came out around the Solaris 11 launch, I'm sure there are many more! Feel free to post a comment below if you find a particularly interesting blog post that hasn't been listed so far and share your enthusiasm for Solaris 11! *Affiliate link: Buy cool stuff and support this blog at no extra cost. We both win! var flattr_uid = '26528'; var flattr_tle = 'Solaris 11 Launch Blog Carnival Roundup'; var flattr_dsc = '<strong>Solaris 11 is here!</strong>And together with the official launch activities, a lot of Oracle and non-Oracle bloggers contributed helpful and informative blog articles to help your datacenter <a href="http://en.wikipedia.org/wiki/Up_to_eleven">go to eleven</a>.Here are some notable blog postings, sorted by category for your Solaris 11 blog-reading pleasure:'; var flattr_tag = 'blogging,digest,Oracle,Solaris,solaris,solaris 11'; var flattr_cat = 'text'; var flattr_url = 'http://constantin.glez.de/blog/2011/11/solaris-11-launch-blog-carnival-roundup'; var flattr_lng = 'en_GB'

  Read the article

< Previous Page | 6 7 8 9 10 11  | Next Page >