HPE EVA Storage
1748035 Members
5149 Online
108757 Solutions
New Discussion юеВ

Re: EVA Perf Counters / Limitations / Recomendations

 
Deano - AIS
Occasional Contributor

EVA Perf Counters / Limitations / Recomendations

IтАЩm currently creating a report for one of our EVA 8100's (fully populated 2C18D config)and cannot find much in the way of guidelines or recommendations / limitations published by HP. Ive had a look across the forum and have found some useful material but would like some more info on the following if possible.

I have run the following counters over a 7 day duration using EVA Perf @ 5 min intervals:-

Controller Status Performance Object
Host Port Statistics Performance Object
Array Status Performance Object
Physical Disk Performance Object

With the Host Port Statistics Performance Object counter what are guidelines for Read & Write Requests & Read & Write MB/s Is there a limit?

What are the guidelines for Read & Write Latency? Am I correct in thinking an average Read Latency < 15ms and Average Write Latency < 5 ms should be acceptable for most workloads?

Do we have guidelines for Av Queue Depth? i.e. 2 is acceptable, 3 is bad etc.

Moving onto the Array Status Performance Object counter do we have guidelines for Total Host MB/s & Total Host Req/s - What do these counters represent? Normal BAU activity for Total Host MB/s is 100MB/s spiking up to 6000 & Total Host Req/s is 4000 spiking up to 20000 - is this acceptable? What are limits of EVA 8100?

With the Disk Group Performance Object counter what are recommendations for Read & Write Latency, BAU seems to be around 17ms spiking up to 37 which seems very high.

What are guidelines / limitations for Read & Write Req/s and Read & Write MB/s.

Any information on the above would be appreciated, I'd like good guidelines and not finger in the air numbers if possible.
5 REPLIES 5
Ivan Ferreira
Honored Contributor

Re: EVA Perf Counters / Limitations / Recomendations

I think that getting physical disk performance is not required as the way eva works.

You should get the physical disk group performance statistics (evaperf pdg). The Read and Write latency really depends of your transactions type. Some hosts are more sensitive to delays than others.

You should ensure that your processor load is balanced between the controllers (evaperf cs).

You should identify the top vdisks and theyir normal load, also, the ammount of data going through the mirror port.

>>> What are guidelines / limitations for Read & Write Req/s and Read & Write MB/s.

This really depends of the model, load distribution, HBAs, SanSwitch, access patterns, etc. What you "could" do is to get your best scenario by running a load/stress tool to get what you could really get on your storage.
Por que hacerlo dificil si es posible hacerlo facil? - Why do it the hard way, when you can do it the easy way?
Deano - AIS
Occasional Contributor

Re: EVA Perf Counters / Limitations / Recomendations

Ivan - Many Thanks for your swift responce however I would still like something a bit more concrete.

The setup I have is HP c-Class enclosures running VMware. 2 x 8 Server clusters per enclosure. Connectivity is BL485 G5 Servers - HBA - mds9124e internal - mds9124e external - (NPV mode) - Core Switch - Storage Port (EVA) - EVA 8100.

Load is evenly distributed across all storage processor ports which the Controller Status Performance Object is showing. 1 controller is working slightly more as the BFS volumes all go through Port 1 on Controller A.

I understand that the controllers should never exceed 45% usage incase of failure the other controller will have to take the additional load etc so controller & host port stats are looking good.

The issue I have is Im seeing some worrying latancy figures, from my understanding and please shout if this isnt correct that an average Read Latency < 15ms and Average Write Latency < 5 ms should be acceptable for most workloads.

Looking at my stats it can be determined that Read Latency is with in recommended guidelines however the Write Latency usually exceeds the recommended thresholds and is aroud 8-9 ms. Is this acceptable for VM's? These VM's are around 1400 Desktop machines or should I put in steps to reduce this figure? If so can you suggest what I might do? Higher speed disks would work but would cost a lot for the benifit of 4 (ms) writa latency.

Any Thoughts or Suggstions would be most welcome

V├нctor Cesp├│n
Honored Contributor

Re: EVA Perf Counters / Limitations / Recomendations

Read latency < 15 ms
Write latency < 5 ms
Host queue depth <= 1

Total Host MB/s = Total traffic sent and received to hosts (this is usually NOT the limit)
Total Host Req/s = Total amount of requests sent by the hosts (this is your problem)

The limits is not imposed by the EVA controllers, they can handle 225,000 I/O Requests per Second, but that's to the cache. The limit are the disks. Each disk can handle 90 - 150 IOPs, depending on what kind of disk and if using VRAID 1 or 5. More info here:

http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?objectID=c01671044

If you're sending 20000 IOPs to the disk group, then either:

1) They are reads and most of the info is cached
2) You have 200 disks on the disk group
3) The EVA is an 4400 / 6400 / 8400 and the disk group is made of SSD disks
4) None of the above, then you'll have a high write latency and host queues above 1

Virtual machines can generate a lot of IOPs, just disabling "Last Modified/Accessed Timestamp" in Windows avoids one write per each file accesed. Also, do not configure all desktops to run antivirus scan or software updates at the same times. And most important, do not use FATA disks to store virtual machines.
Niels Vejrup Pedersen
Respected Contributor

Re: EVA Perf Counters / Limitations / Recomendations

Hello,

Usually you should be fine starting looking at latencies per disk group

Rules of thumb is:
Read miss Latencies max 20ms
Write Latencies max 8ms

(evaperf vdg)

IF you exceed any of these further investigation should be performed to determine the workload causing the bottleneck.

Regards
Deano - AIS
Occasional Contributor

Re: EVA Perf Counters / Limitations / Recomendations

Thanks you all of your replies some of the comments have been very useful - I enclose a link which I have also found to be useful whist creating my baseline report :-

http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?lang=en&cc=uk&taskId=120&prodSeriesId=3664583&prodTypeId=12169&objectID=c01685240

Enjoy