root@raider.UUCP (Bob Reineri) (08/29/88)
Though this newsgroup seems oriented more toward the technical HP systems, :-)
I couldn't find any other group to post in. Please bear with me if I'm slighty
off the subject matter.
I am the system manager of an HP3000 series 70, and my question deals with the
perfomance of disc cacheing on this system.
We are a highly interactive environment, with about 70 on-line users on a
typical day. Much of he work they perform involves Image databases. I am
getting about a 69% reduction in the number of physical I/O's using the
cacheing system but the write hit (ie,process stops / cache requests) is
abnormally high, on the order of 20 percent. I have tried adjusting both the
sequential and random fetch quantums all over the place, to no avail. I can
get no better performance. One thing I did notice is that the system disc,
LDEV 1, usually has about a 100 % write hit rate !
Oh yes, we are running five discs on the system; LDEVS 1 and 2 are 7937
Eagle drives, and the other three are 7935's (404 mb). Any help would be
greatly appreciated.
P.S - IS there a newsgroup for HP's MPE based systems ?
--
______
/ / RaiderNet Public Access - Node 2 (Xenix) (615) 896-8716
/_____/ ___ o ___ ___ ___ * Murfreesboro, Tennessee *
/ \ /__/ / / / /__ /__/ * Middle Tennessee's Gate *
_/ \___/ /__/__/__/__/__ __/ \_ * Node 1 (615) 896-7905 (DOS)*
UUCP: !{ames,osu-cis,rutgers,decwrl,mit-eddie}!killer!raider FIDO:1:116/12Mike_W_Ryan@cup.portal.com (09/01/88)
Regarding the inquiry on disc caching on a S70. Your problem probably stems from the following: 1) locality of data on ldev 1- use LOGAUDIT or HPTREND to identify the files that are hit on ldev alot. Since it is a 571 mb drive this could take some digging. many systems benefit from using a smaller disc on ldev 1 like a 7936 or 7920/25 even to prevent data sets from getting on ldev 1. If you find any suspect files, you can just store them, then restore with the "DEV=" parm. 2) Things aren't as bad as you think. The "official" guideline used by HP to identify systems that are benefitig from cacheing is: a) read hit > 75% b) ratio of reads/write approx 3:1 You are getting something out of caching for sure. 69% reduction is actually pretty good. If you haven't already, you might wan't to try turning BLOCKONWRITE= ON. This usually reduces the number of writes "lined up" behind another write to disc since the writes will then happen at the processes priority rather than background priority at first. Last of all, the way your application performs may be partly responsible, but that remains a constant in most shops.