Overall Performance on P4300 SAS Starter

[portuer]

Hi,

one of our customer have a new Lefthand P4300 G1 installed. We conneted 2x DL380G6 over 2x Procurve 2900 Switches to the Lefthand for redundancy purposes. The new SANIQ SW 8.5 is already installed including the lates patches. We use Sever Virtualization with Microsoft Hyper-V R2 Server (Core). There are (only) 3 VM guests. The latest MPIO Driver is installed.

With the programm HDTUNE (http://www.hdtune.com/) we get reading performance rates from only 30 to 45 MB/s.

Please help!

[Prokopets]

Do you use jumbo-frames?

[portuer]

we don't use jumbo frames.
for now we even do not use flow control.

[Steve McGee]

Are you sure that application is even suitable for testing your SAN?

IMO I wouldn't trust an application designed for testing local HD performance for testing my SAN's performance.

What are your performance monitor counters reading while you're running this test?

[Dave Finch]

HP / Lefthand recommends using IOMeter for performance testing. Be sure to increase the number of outstanding requests. I believe the recommendation is 1x the number of SATA drives or 2x the number of SAS drives in the cluster.

[WishfulThinking]

I'm in agreement with Dave - HDD Tune seems to give poor readings as it's not desgined for external storage.

Give IO Meter a go, I think there's a WhitePaper floating around too somewhere although I cursoru search didn't turn up anything sorry.

[Mario Prinz]

I was wondering because we got far better results by testing an MSA2000SA G1 or an MSA2000i G1 - with the same programm.

[Dave Finch]

Mario, with IOMETER, the key is to increase the number of outstanding requests above the default which is 1. I didn't see an option like that in hdtune from a cursory look at the web site. Increasing outstanding IO requests will increase the Queue Depth which you can monitor in the CMC performance monitor. If the Queue Depth is low, then you are not getting optimal performance on the SAN. It's counter-intuitive, but a high Queue Depth on the HP P4000 is a good thing.

[Mario Prinz]

Thanks for your replies!

I've testet with Iometer with the following settings:

- RAW-Disk 5GB
- 1 worker
- 64 outstanding I/Os per target
- 8KB
- 50/50% Random/Sequential
- 50/50% Read/Write
- 100% Access Specification

I got the followed results (4 tests each):

Host 1:
Total I/Os p. Sec: 965
Total MBs p. Sec: 7,45
Av. I/O Response T: 66
Max.I/O Response T: 1619

Host 2:
Total I/Os p. Sec: 1430
Total MBs p. Sec: 11
Av. I/O Response T: 45
Max.I/O Response T: 3640

What do you say? Should I test with anoter values? Where can I find other results for comparison?

Thanks a lot!!
Mario

[JacobS_4]

We're seeing similar performance on our P4500 5.4TB, 4 node SAS cluster.

The NICS on each node are setup in a ALB configuration and split across two HP Procurve 2910 switches for redundancy. There is a 4 x 1Gb trunk between the two 2910 switches. Jumbo Frames and Flow Control is setup across the board.

The hosts are HP blades with 4 x 1Gb trunks from the blade chassis to each of the 2910 switches. We then have ESX 4 running on the blades and Windows 2008 VMs with Microsoft iSCSI initiators connecting directly to the SAN. Everything works fine, the speed is just poor. We've spoken with Vmware and they've checked our ESX configuration and blessed it. We're not seeing an errors on the switches, but as Portuer noted we also only get about 30 to 45 MB/s. Before we turned on Flow Control and Jumbo Frames we only got 15-23 MB/s. Not quite what we where hoping for based on the investment made.