Read slower then writes on my benchmark

[logicmate]

Configuration:

***Starwind Server
Dual Quad Core Xeon 5620
Windows 2008 R2
12GB Ram
Adaptec 5805Z controller
2 x 50GB SSD (Raid 1 - OS partition)
5 x 600GB SAS 15K (Raid 10 - Storage Partition)
8 Nics

***ESXi Host
Dual Quad Core Xeon 5620
12GB Ram
4 Nics

On the starwind server we did the TCP Tweaks & TcpAckFrequency=1 for every NIC
Network & NICS are configured for Jumbo frames (verified that it's working)
We ran the test inside a W2K8 Virtual Machine on ESXi host (The disk we tested on was created as a VM HardDisk on the RamDisk target, we did not use MS iSCSI initiator)
We are only using 2 iSCSI NICS with MPIO, MPIO is configured correctly as we see transfers above the 1GB NIC limit, MPIO set for RR and IOPS=1
We are testing on a RamDisk target

1. What could cause the read speed to be slightly lower then the write speed?
2. Is there any room for improvements based on the test results?

Thanks,
Ara

[anton (staff)]

You have EXCELLENT performance and perfectly configured software / hardware. There's nothing or very little left to tune. You may play with cache size and policy but it does improve writes more then reads. Tests are synthetic so actual read caching pattern is not exploited.

[logicmate]

I was able to squeeze out more, I think the problem was on my end, I had originally formatted the partition inside the VM HardDisk with 64K allocation size, I reduced it back to the default 4K allocation size and now I seem to be maxing both NICS.

Ara

[anton (staff)]

...and now there's absolutely nothing to tune. Wire speed with both channels. Period.