Which Evaperf counter ?

[Andreag_77]

Hi guys,
i guess to have some I/O performance problem on EVA4000, so i would like to use evaperf for identify possible bottleneck.
I've read Evaperf paper but i don't understand wich (windows) performance counter can i add for my aim.
Someone can tell me some tips or advice?

thanks very much
Andrea

[IBaltay]

Hi,
first of all the Load balancing should be checked: Do you use the ALB or SLB?
ALB=MultipathIO (DSM 2.X, 3.X) on the hosts with (SQST LB setting) and the EVA Set preferred path = NONE - this is recommended as the optimum for most windows apps i the Active/Active EVA.

[Andreag_77]

thanks, but where i can find these EVA informations?
I've seen command view but i don't find it.

thanks very much
Andrea

[IBaltay]

Hi, you can see it in the presentation of each VDISK

[Andreag_77]

ok, i find these info on all vdisk:

OS unit ID: 0
Preferred path/mode: No Preference
Managing controller: Controller A

SCSI Reservation : None

All vdisks are in RAID5, but no other relevant informations in presentation of VDISK

[IBaltay]

Hi,
and are you using the MPIO load balancing pls? If not, then the host<->eva communication is not balanced which means that the whole communication can go only via one controller.

[Andreag_77]

where i can find if it there is MPIO load balancing configured on my EVA?
I've seen all Command View configuration but it doesn't.

thanks again

[IBaltay]

Hi, you must go to your win host and there it is accessible via the HPDSM CLI.

[Andreag_77]

unfortunately in Command View windows host there isn't any hpdsm utility :(


However, you can tell me which data i have to verify with Evaperf for identify possible performance poor?

thanks!
Andrew

[IBaltay · Accepted solution]

Hi,
1. the basic Performance metrics:
array stats (evaperf as)
controller stats (evaperf cs) -
host port stats (evaperf hps) - IO latency
physical disk group stats (evaperf pdg)
you can run this 4 in a separate CMD windows and you can see the overview of the perf situation

(combine it with the -csv operand to import the data to e.g. any perf grapher or excell itself)

2. then according to the bad host port statistics you can run specific
vdisks performance data evaperf vd -fvd

The key performance parameter:

1. IO latency

values:
a) read miss latency
<10ms =ecxellent
10<20ms =very good
20<30ms =acceptable
30ms and more =unacceptable
b) write latency
<5ms (with write cache enabled) =ok
5ms and more =unacceptable,
we need to check the physical disks IO request queuing (evaperf pd, pdg)
>1.5-2 queue depthts - not enough spindles in a disk group (add more physical disks)

2. Host queue depth (hc)= io requests at a time
example: the default queue depth per host HBA for the windows server is 32. If the actual values are less then potential workload, then we can say that the host is not giving the array enough work to do. If it is equal/similar, the HBA server side queue depth settings should be tuned