meaning of "other " latency

Question

Hi,

I just installed SRM on my Orion server. I added a NetApp running clustered mode and start to notice some of volumes have very high latency (instead of a few msecs, they are ~200msecs). Only some of the volumes on the NetApp have such high values and the NetApp is on the same network as the SRM server. Drilling down to the latency graph of one of the volumes, it shows:

Total: 182 ms

Read: < 1ms

Write: < 1ms

Other: 195 ms

First of all, what is "Other"?

Second, the numbers do not add up. Shouldn't "total" > the other numbers?

Any suggestions are very much appreciated.

Thanks,

LubomirKrausko · Answer

Hi,

well, my knowledge about vendor's tools like that is minimal. What I am able to see is the metric in their API, but that information will not help you

Regards

Lubo

acmon · Answer

Lubo,

Thanks a lot for the detailed explanation. I think I have a much more clear picture now.

To chase down the "Other" latency, is there a way to find out which NetApp performance counter SRM polls from? In OnCommand Performance Manager, I don't see any relevant metric corresponding to it.

Thank you.

LubomirKrausko · Answer

Hi,

so in general SRM does not do any measurement nor load on the device itself. SRM only reads counter/statistic values that are on the device itself. It means that it does not matter whether the device and SRM are installed in a one network, the latency SRM displays is only number the device reports with no relation to network.

By these words you can also see, that SRM does not increase any load not traffic on the device itself, SRM only reads statistics from it.

Now the values you are seeing...

The Other latency is metric, that reports device directly to SRM. It only takes it and displays together with other metrics. So why it is so high is better to find the answer in device itself.

Is not the Total latency SUM of Read, Write and Other latency? Actually not. In general if there is a volume that reports Read latency very small and Write latency quite big, the Total latency cannot be the max value, and also it cannot be summary of those two. It should be some value that informs end user about how the volume responds based on the R and W latency but also I/O must be taken in account here. The Total latency here means "kind of" weighted average of all three metrics (Read, Write and Total) where it takes the particular latency metrics the device reports and as a weight is used the Total I/O operations. Not saying it is exactly weighted average, the formula is more complicated, but it can be very close to it and can be used for illustration.

I do not want to share here a lot of details of that, since this data reporting and metrics availability for each vendor can differ a lot, but I hope I did not confused you too much and brought you some level of information you'd like to know about this.

Thanks you for this question and interest!

Regards

Lubo