Hi all,
I have DL380 Gen8 server with P2000 iSCSI array - dual controllers, 1GB version. iSCSI traffic is going through separate switch, with 2 server ports and all array ports. Using ESX 5.5 build 2302651. I followed HP best practices and tons of guides, using RR with iops=1 (but tested also other options). iSCSI setup is tested several times against VMware/HP whitepapers and it works, as well as load balancing, confirmed using esxtop.
Currently I have test Centos 7 machine, with tuned-adm applied, XFS layout. Testing with iozone, no matter what and I how I change, I'm getting sequential reads close to max 2x1GB interfaces: 245MB/sec, but read performance is maximum 150MB/sec. I tested various disk layouts on P2000, the best seem to be RAID10, but interestingly it doesn't change much between RAIDs, I have pretty much the same write performance on RAID5, just worse read, down to even 20-30MB/sec. Surely, P2000 with 12 10k SAS disks should be better than this.
I've spoke with HP support, they've seen array logs and they say it's all good over there and it's rather VMware problem, but honestly I can't find it and I'm out of ideas. At the moment I'm using RDMs, they are somewhat faster, with VMware paravirtual scsi controller. I also tested Centos 6.6 and Windows 2012, in fact Windows is even worse than other two.
Any hints from you guys what might be wrong with my setup?
Many thanks
Chris
