I have Centos 6.2 in server with this specs:

2xCPU 16 Core AMD Opteron 6282 SE
64GB RAM
Raid controller H700 1GB cache NV
- 2HD 74GB SAS 15Krpm RAID1 stripe 16k (OS Centos 6.2) sda
- 4HD 146GB SAS 15Krpm RAID10 stripe 16k (ext4 bs 4096, no barriers) sdb -> /vol01
Raid controller H800 1GB cache nv
- MD1200 12HD 300GB SAS 15Krpm RAID10 stripe 256k (For DB Postgres 8.3.18) (ext4 bs 4096, stride 64, stripe-width 384, no barriers) sdc -> /vol02

I'm benchmarking IO speed with dd, and view thah if in RAID10 12 disk exec:

dd if=/dev/zero of=DD bs=8M count=10000 oflag=direct
10000+0 records in
10000+0 records out
83886080000 bytes (84 GB) copied, 126,03 s, 666 MB/s

but if I remove "oflag=direct" option obtain about 80 MB/s. In read benchmark, results are similar:

dd of=/dev/null if=DD bs=8M count=10000 iflag=direct
10000+0 records in
10000+0 records out
83886080000 bytes (84 GB) copied, 79,5918 s, 1,1 GB/s

If remove iflag=direct obtain 150MB/s...

I don't understand this huge differences, on other machines y don't have this behavior.

Can I have some kernel parameter misconfigured?

Thanks!