Cluster restripe times

[Gediminas Vilutis]

Does anyone have some ideas how to calculate cluster restripe times when adding/removing storage nodes from it?

The problem is that sometimes we need to remove/add storage nodes from/to cluster and need to know how long this restripe will continue (in order to make sure that nobody will start intensive IO during restripe). HP support is completely unusefull and clueless - 'restripe can take from day to week, blah blah blah' :(

So question is - say, we have x TB provisioned in cluster of N nodes (i.e. x/N TB per node). 'Local bandwith' setting for management group, e.g., 40 MB/s. How long it will take to restripe cluster if you add 1 storage node? Any ideas?

Gediminas

[oikjn]

you are one your own for that one.  Only thing you can know for sure is the MINIMUM time as that is regulated by your set bandwindth limit.

What I've just done is turn the bandwidth down to ~5MB/s to slow down the restripe rate to not effect production performance.  Also, when the speed is set that low, you can be more sure that the actual time of restrip will be close to the size of the data divided by the speed limit set in the management group.

I've never been able to slow down production to do a restripe and don't see it happening unless we have a major cluster blowout and redundancy is lost and we need to get it back ASAP, but that is not the case when you are doing planned node changes.