February 28, 2014
How MPE Balances New Disk Space
If we have a system (volume set) with mostly full disks, and I add a new big empty disk to it, how will MPE/iX do all new allocation on that disk — will it wait until it fills up to the same relative fullness as the existing drives?
See, we have a system with a Nike Model 20 and a bunch of RAID 1 LUNS, and we’ve added five new drives in RAID 5 to the system volume set. But that sounds like we’re on the cusp of a disaster, because while the read performance is measurably better, all the system is going to be doing is writes to this drive for every new extract and scratch file. And as everybody knows, the write performance is like 2.8 times slower to the RAID 5 LUN than the RAID 1 LUN.
[Corrected, to identify the BALANCE command as a part of DeFrag/X.]
Craig Lalley noted, "There is a command you will want to use if you have Defrag/X, [created by Lund, sold by Allegro] The command is BALANCE VS. As an example,
"There’s online help for this command in Defrag at HELP BALANCE. Without that, I would use system logging to determine the most heavily accessed files and store/restore them to spread the extents."
And there's also help to manage this kind of balancing and defragmentation from VEsoft, as well as that Lund tool.HP designed the 3000's storage management so that the MPE/iX algorithm will be picking which "most empty" disk to write to next based on percentage full, not sector counts.
MPE watches the percentage full that will cause a switch to another disk. The watching has gotten more precise. An older algorithm would wait until there was a 1 percentage difference in fullness to switch -- and as an example, that would mean 3 GB of data on a 300 GB disk. Now MPE waits until there’s just a .01 percentage.
MPEX from VEsoft can be an aid for manual load balancing after disk installation, before production use. It has commands that can be used to build a DEFRAG process. There's also that DeFrag/X software from Lund to manage storage assignment.
As a last resort, one HP MPE veteran suggests that if nothing else helps, and no MPEX or DEFRAG are at hand, "managers could try to limit the penalty by moving large but less "interesting" files/accounts to the new LDEV (freeing space on the old LDEVs) and then use the VOLUTIL ALTERVOL command to limit the remaining PERM space on the new LDEV afterwards. Yes, a somewhat insane approach, I know."
Use our search engine to find 20 years
of HP 3000 news and articles
The comments to this entry are closed.