Volume Example Performance question?

[romatlo32]

Hi folks,

I am being asked to verify that storage volume performance was good while a VM's workload (Oracle database query) was shifted from VM memory to VM disk.

There was a bug reported that caused the workload to shift to disk.  It's expected that the workload would be slower on VM disk than memory.

The complaints and performance was so much slower that we were forced to reboot the server, which cleared it up and now running in VM memory again, no problem.

I ran a performance report on our Nimble CS215 for that volume (only the one VM using it).

The slow performance was reported from March 12-15.

I see more read activity in the 3/12-15 timeframe (expected when workload shifted), but do the numbers look bad?

Anyone see anything that could cause concern?

I would like to tell our management that if this happens again, our storage is optimal...

[sdaniel47 · Accepted solution]

A quick glance at your graphs shows nothing wrong with storage.  Your cache hit rate stays very high and your response time stays low.

This suggests a couple avenues of investigation:

• One possibility is that switching from memory, which responds in tens of microseconds, to reading from cache on the array, which responds in a hundreds of microseconds, is a big enough slowdown to cause the problem.  No matter how fast your array, it is still 10 to 100 times slower than memory.
• Another possibility is that you have a problem higher in the stack than the array.  Some problem with the network or the way the server's iSCSI is configured is preventing you from getting more performance from the array.

My recommendation is that you open a support call with Nimble and ask them to help you trouble-shoot the issue.

One last question: do you have or can you produce Oracle AWR reports from the period when things ran well and the period when things did not run well?

[romatlo32]

Thank you Stephen!

I will check on the AWR report and reply back, hopefully close this thread early this week.

[Valdereth]

Looks solid - the latency remains unchanged even when the disk I/O increased.  Keep in mind if the application is maintaining its own cache in VM Memory that will probably keep some data from being placed in SSD cache, so expect a brief ding in performance (latency) when the switch to disk occurs and initial reads are served from HDD.  After that Nimble will serve the cache worthy data from SSD and you'll be in good shape.

Further tuning may be available, but I'd be surprised if you saw any significant jump in performance.

[romatlo32]

Thank you folks for responding.  Loving this Nimble!