WLA Performance thresholds

[villavic]

I have Symmetrix 8830 and DMX, san fabrics (2GB switches, though some HBAs at the servers are still 1GB).

Some CCenter alerts by default send this message:
AlertDescription:Symmetrix:KB/secreadis417495 (critical)

And when I take a report of WLA I can see high peaks of throughput, but I don't know if those peaks are actually a problem.

How can I be sure how much KBsec are too much (read or write?). It's easy to know the %hit, %writes, etc, and find out overloads. But in terms of numbers, throughput, how can we be sure how many frames, how much KB are too much? Is there a rule of thumb? tips? guide?

 

[Chacalinc]

there are no rules of thumb for this kind of issues. It depends on your environment. If you see that some ports have too much I/Os abd others are more free, then you could redistribute the ports to the differents servers. Alerts are defined in the WLA but this is defined for every environment and an alert for you could be a normal I/O for another company...


Cheers.

 

[culainn]

I agree with Chacalinc on this one. Also, interface with the vendor on this one. They should be able to provide you with some statistics i.e. IOPs on port. The thing you need to understand is the profile of your workload and what blocksize it is using, as blocksize is a determining factor in the number of IOPS you push through a port and and also the throunput (MB/sec), i.e. at a low block size circa 4KB you may see alot of IOPs but your throughput would suck, conversely, larger blocksizes would see low IOPS but high thoughput. What I am getting at is that you may see loadsa of IOPs going through a port and you think you are doing well, in fact you may have a low blocksize and thus throughput. A true idea of performance is built up over time, with historical data you can perform comparitive analysis. All I will say is profile you environment, know what is good for your environment. The devil is in the detail.