How you explain these performances?

[voodooFX]

LSI 9260-8i

Case 1
5x Seagate Constellation ES 2.0TB (+1x dedicated hot spare)
RAID 5 - Strip Size 256KB - Write Back enabled

read/write performance: 300~325MB/s
no way to get more

Case 2
6x Seagate Constellation ES 2.0TB
RAID 5 - Strip Size 256KB - Write Back enabled

read/write performance: 600~640MB/s

For testing I'm using dd in this way
write: dd bs=1M count=1234 if=/dev/zero of=/mount/point conv=fdatasync
read: dd bs=1M count=1234 if=/mount/point of=/dev/null

I can not understand these numbers, why 5 drives are so slow?
Since the Strip size is 256KB shouldn't 4 drives (4+1) be the optimal situation (1MB/4=256)?
Why I'm not getting something around 500MB/s with 5 drives? And why I'm getting 600+MB/s with 6 drives and he's awful 1MB/5= 204.8 ?

INFO
The controller is directly connected to a ESXi 5.5U2 host
Tests are on a debian 7.8 VM with 64GB Tick (Eager Zeroed) disk

 

[Jeggs101]

Totally stumped if they are indeed setup like that.

 

[voodooFX]

To be sure 100% I have deleted the 5 drives volume and recreated the 6 drives volume

This is what I'm getting with the background initialization still running!

root@deb01:/tmp# dd bs=1M count=6072 if=/dev/zero of=/tmp/speedtest conv=fdatasync
6072+0 records in
6072+0 records out
6366953472 bytes (6.4 GB) copied, 12.0334 s, 529 MB/s

 

[mrkrad]

Google Translate

here's a website comparing # of disks versus strip size versus ntfs size versus speed fyi

 

[voodooFX]

thank you mrkrad but it's not really helpful
this guy is using the intel "fake" raid in windows and even with a bug when configured with 6 drives...

my "problem" is that I have very good performances with 6 drives and very poor with 5 drives (which should be the optimal number of drives) and I can not understand why..

 

[mrkrad]

Try adjusting the 5 drive strip size down!

 

[Patriot]

Run collectl to see how the OS is actually handing down the write to the drives. also... try the -direct flag.

Maybe consider running FIO to get async performance too...