EVA Performance

 
Palima
Occasional Advisor

EVA Performance


Hello every body, I'm new to this forum and also to SAN environment so if my question is already answered please just link me there,

We have EVA5000 storage which connected by 2 nodes of trucluster and windows server,

I need to have an idea about the performance, I need to know how much transfer rate I'm using and whether we will be in a bandwidth bottleneck, if we duplicate that rate, or not.

It is an info question and I appreciate detailed actions as I'm still beginner-mid level regarding the SAN environment.


Thanks

6 REPLIES 6
Víctor Cespón
Honored Contributor

Re: EVA Performance

In the server where you have Command View, you will alse have an utility called EVAPerf. It can be used in two ways:

1) Command line
2) Through the counters it adds to the Windows performance monitor (perfmon.exe)

Check this document:
http://h71028.www7.hp.com/ERC/downloads/5983-1674EN.pdf

You can also set it to capture data for a period of time, logging to a csv file. Then use TLVIZ formatter to sort the data chronologically so you can display it on an Excel chart.
Palima
Occasional Advisor

Re: EVA Performance

Thank you very much

I have read quickly the document that you refer me to, in addition to this helpful one http://h10032.www1.hp.com/ctg/Manual/c00605846.pdf

I used counters on windows performance monitor and access database, I can see that the controllers Cpu's usage only reach up to 15% so I guess I have no problem neither I will have after the load duplication in controllers, true?

Also I checked the HP EVA storage array â total host KB/s and found it's peak is 82 MB/s so I guess that the maximum throughput used is 82*8 Mb/s = about 650Mb/s, sooo we are using about 30% of the maximum throughput which is 2Gb/s,
Is this the way that we forecast the eva5000 storage performance? Am I right? Do I need to check more counters to find if I will be in a performence problem after duplicating the load?

Also when we use monitor utility on tru64 we found that some disks have queue more than 12 instantly so what I need to check in order to fix it and the most important is what I need to do to avoid it when doubling the load?

Thanks for your help and this time I hope to have more specific in point answer ï as I'm new to this it hang me few days to read and understand the document, may be still not ï

Thanks
Víctor Cespón
Honored Contributor

Re: EVA Performance

The numbers to watch are always the latencies.

- Read and write latencies on the disk group
- Read and write latencies on the controller ports
- Queue length on the hosts

The controller CPU is never a limit
The FC port bandwith usually is not either (you have 4 FC ports at 200 MB/s each)

With mechanical disks, the limit is usually the number of I/O operations. Anyone with experience copying files on a computer, to any kind of storage device knows that it's not the same copying 1000 * 4 KB files than a single 4 MB file.

Read latency should be below 15 ms, write latency below 5 ms. Queue depth on the hosts 1 or 2 as maximum.

If they are higher then each port, host and vdisk must be examined individually to see which ones have the most traffic.
Palima
Occasional Advisor

Re: EVA Performance


Hello Vcespon

Thanks for your response and sorry being late for another problem,

I used performance monitor and added physical Disk group counters and the reading of the latencies is as follow: for reading the maximum reached in some ports is up to 35ms and the average is 7ms and regarding the write latencies the maximum also exceeds the 5ms but the average is about 2ms.

Regarding the queue depth: I found three counters oblects as follow:
- Under the HP EVA host connection object there is a counter for queue depth and the reading average is 1 for most of the host ports sometimes it reached 2, this is the average, I noticed that some counters jumps occasionally up to 24 value and get back again to normal,
- Under EVA host port statistic object there is a counter for Av queue depth and we have 4 ports and the average for these ports are between 6-12 and the maximum sometimes reached 30
- Under Physical disk drive there is a counter of drive queue depth and the reading is 1 as average and 2 as maximum,

I think this is fine if your numbers means average but I don't know if doubling the I/O will double the latencies value? In that case I will be in problem, true?

I assigned 5 point to yur previous answer because it is really useful and I think I can assign more for new answers and I hope it is fair,

Thanks again and waiting your comment regarding effect of doubling the I/O load on the latency value.

Víctor Cespón
Honored Contributor

Re: EVA Performance

Check these two articles:

Basic EVA Performance Guidelines
http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?objectID=c01685240

Relative Performance of Disks Used in the EVA
http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?objectID=c01671044

Look at the article above, multiply the number of I/Os each disk can sustain by the number of disks on the disk group.
When you go over that limit, the latency goes above 15 ms for read and 5 for write very quickly.
Palima
Occasional Advisor

Re: EVA Performance

I have got usefull help

Thanks man