Working out data skew from the hosts

Hi

Could you define what you mean my 'skew'?  Might be able to help if I knew that you are wanting to achieve.

I'm quessing that from Fast VP - you would like to work out number of drives per type to add to a pool?

Phil

Sorry about the delay! What I mean by Skew is the percentage of capacity that is producing the majority of the IO in the environment. EMC use this term when sizing VNX with Fast Cache and/or Fast cache. However using lun level skew isn't effective as when I have looked at VNX or Claiirons both at lun skew and sub-lun skew the results are very different.  The issue I have is sizing VNX with the fast technologies without an idea of the sub-lun skew in the environment as in some cases the IO is fairly evenly distributed across the capacity. This means that some active data may find its way to the NL drives and cause response time issues. I have tried to come up with a spread sheet to calculate skew based on ratios i.e. lun IO / capacity but without the best results. I maybe asking for too much with current tools but I thought I throw it out there!

I dont think you can reliably do this – calculate information from sources that don’t contain it.

To get reliable information on skew you would need to have stats on which blocks were accessed.

I don’t think any client tracks that

Nor do the Analyzer NAR files

The only reliable way is to collect RBA traces on the VNX – these contain block number and timestamp for every I/O

From there some EMC internal tools can make a good simulation

Thanks for the response, I guess I was asking too much really. I'm not using rule-of-thumb sizing based loosely on lun level skew with the caveat that one or more of the tiers may require upgrading depending on the real-world skew.

Worth noting that Mitrend now includes a skew assessment for systems running 32+ this is due to a new naviccli command being added.