I have so many questions and I don't know where to start ....
Something sounds very wrong if you have such clear hot-spots on the physical layer. The 3PAR should distribute data pretty good across multiple LDs crossing multiple PDs when writing to a volume. The only thing that might prevent this is if the data set if very small. In that case it should hit cache and you shouldn't get hammered on the backend.
Alternatively you could have created a CPG and filtered it to only cover 2 or 6 PDs but again, that would be against best practice and just plain stupid.
If this is constant I would try looking at the 3PAR CLI reference guide at the stat commands. statpd you clearly know, statld will point you to which LD is getting hammered, statvv will show which volume and statvlun could point out which server.
srrgiodensity could help you seeing how big the hot data set is.
_________________ The views and opinions expressed are my own and do not necessarily reflect those of my current or previous employers.
|