HPE Storage Users Group

new registant here with what is probably a quick technical question I've not seen explained elsewhere (I don't have a 3par yet).

Are 1GB chunklets (in a 7000 series) raided into only with chunklets from the same size drives or can drives of different sizes (but the same "type" - e.g. fc/nl) be used together.

So if we bought an array with 900GB SAS drives and later something bigger comes out would there be one or two allocation pools within the CPG - one with chunklets from the 900GB disks and one from the 1.2TB drives say or do they all get mixed together since they are all 1GB in size?

Logically I would presume they are all mixed - just some disks have fewer chunklets than others.

The chunklets are grouped by whatever drives you put in the CPG. So if you have 900GB drives today in a CPG and then add 1.2TB drives later and add them tot eh same CPG then chunklets for a given volume will be spread across different drive sizes. We faced a similar delemia last year. We bought our array with 300Gb 15K drives and 2TB NL drives. We needed more faster drives, but 300GB drives were not cost effective. We bought 900GB 10K drives (around same IOP per drive). 3par considers 15K FC and 10K SAS both as FC, which conveniently now means "Fast Class" instead of Fibre Channel. We decided in order to maintain balanced data on drives and not be lop sided on drives that we would add all the 900GB 10K drives to their our CPG so that data would spread evenly across drives of the same size. Of course that that also means is your workload and IOPS are now restricted to what is in the given CPG. If you are not worried about data being uneven tand the drives offer same IOP profile then you can mix them, if you want to have data land more evenly and the aggregate IOPS is not a factor then seperate CPG is better.

Yes they can be. It all depends on your CPG settings. You can setup very complex filters and criteria in a CPG, although its not recommended.

My CPGs appear to be locked down by drive type and speed , FC 15K, so any capacity FC 15k drive would be used by my CPGs.

The CPG' are just policies so you can have as many or as few as needed e.g

CPG1 = 900GB only
CPG2 = 1200GB only

The above provides separation, but limits max performance, it also complicates choices when you need to add more IO or capacity.

The below would provide the best performance, however once the 900GB drives are full all new writes can only land on the 1200GB drives so dependent on the drive numbers you may see a tail off in performance. However your upgrade path is now much simpler, just add 1200GB drives going forward, whether for capacity or performance. Also over time as you add more the larger drives will predominate and provide relatively more IO.

CPG3 = 900GB & 1200GB combined

If you want flexibility you can have all three, but then you have to think about where best to place data and also factor in the 900GB only CPG will run out of space at some point. Bottom line, you'll get best performance and simplest management by aggregating IO across all spindles of given class.

Thanks for the answers guys (?) - that's what I thought - obviously sticking to 900GB is best as otherwise the 1200GB disks get asked to do "more work" (since they have more chunklets and are part of more LD's) which isn't best for performance.

We haven't even switched the 7200's on yet and I don't feel we are going to be short on space - but more I/O spindles might be needed in due course if performance isn't reaching the desired figures - or as is more likely the workload increases beyond where it was designed to be.

There's no technical reason to segregate the drives, it's more of an individual management or internal political decision. You can separate them but then you have to begin planning which drives you need to add where to support individual workloads, a bit like a traditional array where you have discrete raid groups / disk groups / aggregates etc. That's typically where the desire to physically partition things comes from. Since the capacity differential between 900GB and 1200GB is minimal in reality you just need to keep an eye on the physical capacity consumption on the smaller (see System reporter daily reports) drives and plan ahead.

Chunklets are distributed evenly across all drives of a given class (FC) regardless of their capacity. Meaning the 1200GB drives will only have to do more work once the 900GB drives are 100% full, so as stated above aggregating both 900GB & 1200GB in the same CPG will provide the best performance. In a combined CPG performance is spread across all disks up to the point the 900GB drives are full, at which point the 1200GB drives will service net new writes (not necessarily overwrites) and over time a higher proportion of reads vs the 900GB drives. Add drives of either capacity to the CPG and you have additional capacity (relative to drive type) and more performance (equivalent IOps), both are 10K RPM anyway.

Since data is wide striped, performance is aggregated across all spindles, so rather than trying to place data in a silo. You would be better just adding more 1200GB drives going forward, which is also a no brainer from a planning perspective. With this model over time you will also end up with more 1200GB spindles anyway, meaning they will provide better aggregate performance than the original 900GB drives would have on their own.

One thing to keep in mind given the drive sizes you're dealing with and possibly a reason to separate, is whether you need to consider the use of raid 6 on the 1.2TB drives. In which case you could go raid 6 across a common 900/1200 CPG or filter with raid 5 on 900 and raid 6 on the 1.2TB.

BTW if you decide you want to change any of this at any point in the future, assuming you have some space available, you can simply re-tune to a new CPG layout with DO.