Determining workload type (random versus sequential)

Question

Forgive me if this is obvious, but is there a way to determine the type of workload a lun is performing from the array itself? I know I can do it via Windows Performance Monitor for certain database apps ( if Avg. Disk Bytes / Transfer is significantly higher than 8 KB, for example), but is there a report or analysis I can do on Celerra (or Clariion, for that matter) to tell me this? I have a specific customer with several custom applications and no 'application developer support' on any of them. So I'm completely in the dark with the apps themselves and I have a feeling the load is sequential, not random, and therefore suffering in its current configuration on the array (mixed random & sequential workloads on the same luns). How do I confirm this (I don't have the luxury of trying and seeing what happens. The case for a change needs to be made first)? Thanks for any guidance you can provide. John p.s. before you ask, NaviAnalyser shows some issues, but I need to be able to go deeper.

Rainer_EMC · Answer

you'd have to do that on the Clariion with NaviSphere Analyzer - the Celerra doesnt capture that much I/O details

I dont know a way to go deaper than Navi Analyzer - you might want to ask in the Clariion Forum

jcrsi · Answer

Thanks Rainer, I will do just that. Cheers

BillStein-Dell · Answer

Remeber too, a workload that looks sequential to the Data Mover may not look sequential to the LUN. Also consider how the backend is laid out. How many filesystems share that RAID group? Is it more than one, and are they accessing the same RAID group at the same time? Sequential data being presented to the LUN can actually produce the same I/O characteristics of random I/O if there are multiple I/O streams hitting that RAID group.

In order to fully understand whether or not a workload will show performance characteristics of either sequential or random I/O, you have to understand:

How the workload looks to the Celerra
How the Celerra views the LUN (i.e. a filesystem, a virtual tape, etc.)
If it is a filesystem, is it striped/sliced across disks? Do the slices overlap? Could the distribution of writes loop back such that you'd be hitting two segments on the same LUN?
The RAID type in relation to the workload (i.e. is the RAID type appropriate for the workload)
How many filesystems share the same RAID group

Hopefully I didn't muddy the water, but when you are talking about performance characteristics when a Celerra is involved, there are several layers that have to be considered. It's not as cut and dried as direct access to the storage array is.

Celerra

Was this post helpful?