RE: Multiple LPARs PowerVM and internal drives

I don't know the size of your system nor your HA plans. However I will say this...
We have a system with only 4 drives in the system unit. One pair mirrored for one vios lpar. Another pair mirrored for the second vios lpar.
It has one lpar of IBM with 23TB of internal SSD's in expansion units. That lpar is a hosting lpar only and hosts the disk for four other lpars of IBM i and one of AIX. Using the hosting lpar concept allows all guests to use all the disk 'arms' and not have to throw money at numerous disk controllers per lpar.

Last week I had IBM connect and do a performance lpar on the main guest, runs INFOR LX for our ERP. They pointed out a few issues but in general there were only two, very minor, issues with disk. One was there was too much temporary space which can impede the use of SQL plan cache. They suggested more frequent IPLs. I plan on using the DB2 services for analyzing temporary space instead.
The other disk issue needs a little background information for those unfamiliar with IBM i hosting IBM i disk. On the hosting lpar you configure one or more NWSD's (Network Storage Device) per lpar. Each NWSD contains 1-16 network storage spaces. A network storage space is stored in the IFS and is automatically spread across many disks. This is how each lpar shares all the disk arms. The guest lpar they analyzed has four NWSD's going to it. One NWSD is only for devices other than disk. The other 3 NWSD's host 10, 11, 9 storage spaces. Using multiple NWSD's and multiple storage spaces can be used to help performance. The guest thinks of the storage spaces as disks and you actually go into disk stuff in SST to add them. Thus it thinks of them as arms and adjusts performance accordingly. The guest thinks of the NWSD's as individual disk controllers. Sometimes you don't create multiples initially but you do it as the system grows. Initially most of our stuff was due to growth. But after a few system upgrades we've gotten a better handle on it and preplan for multiple storage spaces spread out over multiple NWSD's. Even so, IBM has determined that one of those NWSD's (which they saw as a disk controller) is busier than the others and would benefit from running the TRCASPBAL and STRASPBAL TYPE(*USAGE) pair.

So, in summary, we use internal disk and host 7.8TB to the guest IBM analyzed and it's running pretty fat, dumb and happy. Slinging a pot of money at external SAN may not be a magic cure all for performance.
There are a lot of other arguments for using SAN outside of performance.


Rob Berendt