MPIO no performance boost?

[Zeon]

Hey guys,
I've just started investigating MPIO with Starwind and haven't really seent he performance improvements I was hoping. It seems even with MPIO I can't get past 1gbps, even with multiple VMs!.

Config:
Starwind
win 2008 r2, 6x 3TB drives RAID 6 with SSD cache and 1GB write back on LSI card. 2x intel NIC for iSCSI, 9k jumbo frames. Raw performance:


Switch
Cisco SGE 2000 with jumbo frames enabled


Host
VMWare esxi 5 u1 on new Supermicro Xeon e5 system with dual intel NIC for iSCSI. Connects to single target on Starwind:







So here is my VM performance. Also in the 2nd picture see that the Starwind server is only using 50% of the bandwidth on each connection?




Any ideas?

[anton (staff)]

1) Start with dumping your benchmark tool. Use Intel I/O Meter.

2) Please list StarWind config. How many nodes, device type, amount of write back cache allocated etc.

3) Please run I/O Meter against local storage and then again from a few VMs you're using to get combined numbers.

[logicmate]

Have you set your IOPS=1?

Ara

[Anatoly (staff)]

I'm assuming that you mean Number of Outstanding IOps, correct? The value should be 32 or 64.

[logicmate]

No I'm referring to the Round Robin policy in iSCSI initiator and settings IOPS=1, default is IOPS=1000.

http://www.starwindsoftware.com/forums/ ... ops#p12194

I'm assuming that you mean Number of Outstanding IOps, correct? The value should be 32 or 64.

[Anatoly (staff)]

Oh! Good question, logicmate!

Zeon, would you be so kind to answer the question, and probably provide us the screenshot that would showed the value of the corresponding setting?

Thank you

[Zeon]

Hey guys,
Thanks very much fort he advice so far. I have run some tests using IOmeter with the following settings:

size: 4096
% of size: 100%
% read: 67
% random: 100
delay: 0
burst: 1

This ran 4x tests. I have given the raw performance of the SAN, performance of 1 VM running the test before the IOPS=1, 2x VMs running IOmeter concurrently before the IOPS =1 and a test after changing the IOPS=1 for a single VM.

For random tests the performance drop because of iSCSI doesn't seem to be too huge. I'm wondering if I should use a different ICF for iometer to perhaps measure sequential read/write?

IOPS


Read Performance (MBps)



here are the starwind settings:



Write Performance(MBps)

[Anatoly (staff)]

Well, you still haven't mentioned the Number of Outstanding IOPS that you have used. As you can see in my post above I've recommended to use 32 or 64 value.