XP24 with External Storage EVA8K performance problems

[name_27]

I have multiple host connected and few of them are really slow disk i/o (0,3MB/sec). With similar configurations some are not (30MB/sec).
Path parameters and CLPR are set ok.
EVA LUNs are presented failover/failback
I'm wondering about 8 paths to EVA causing this?
EVACtrl-A -> XP
500...48 CL3-E
500...49 CL4-E
500...4A CL3-E
500...4B CL4-E
EVACtrl-B -> XP
500...4C CL3-E
500...4D CL4-E
500...4E CL3-E
500...4F CL4-E

XP priority:
1 500...48 CL3-E
2 500...4A CL3-E
3 500...4C CL3-E
4 500...4E CL3-E
5 500...4D CL4-E
6 500...4B CL4-E
7 500...4D CL4-E
8 500...4F CL4-E

Any advice?
Thanks,
-T-

[krusty]

Hi "T",

First, HP has a handy White Paper about External Storage on the XP: http://h71028.www7.hp.com/ERC/downloads/4AA0-6162ENW.pdf

Next, you may want to look at utilizing additional ports on the XP. Currently you are running all 8 EVA fabric ports thru 2 ports on the XP. Those ports -may- be saturated. Typically you should spread the load a little more. Note on the XP, the different CHA ports can share microprocessors. For example, CL3-E uses the same MP as CL7-E. Likewise, CL4-E and CL8-E share MP's. See if the "buddy" ports are being used also.

You really need a detailed analysis of the layout, but maybe this would help.

Cheers,

Curt

[IBaltay]

Hi,
you can also monitor the CHIP load via the PAXP (Performance Advisor)

[IBaltay]

Hi,
here are some of the key ES recommendations:

a) ES XP Cache mode - cache disabled
b) put ES XP in a separate CLPR
c) up to 4 ES XP links to the same AA EVA
d) at least 2 XP links per EVA for HA reasons
...

[Matthew Secaur]

IBaltay,
I am in a similar situation as the original poster (except that our XP -> EVA8000 is using 8 XP ports). You said:

"here are some of the key ES recommendations:

a) ES XP Cache mode - cache disabled
b) put ES XP in a separate CLPR
c) up to 4 ES XP links to the same AA EVA
d) at least 2 XP links per EVA for HA reasons"

I understand all of them except letter "c)". Why only 4 links? If you look at the original poster's "XP priority" table, I have that same config times 8 (That is, 8 path groups of 8 paths, each using a different "primary" path). Is this too many? Could this be the source of my performance problems?

[IBaltay]

Hi,
because in EVA it is always only 1 controller in the role of the managing controller, that means only 4 FPs from 8 is directly accessing the EVA disk subsystem (especialy reads), the other 4 of the second controller - proxy are accessing the disk subsystem only via the mirror port channels between the 2 controllers which is not optimum and could become the performance bottleneck.