Re: FW: Latch Waits

I was talking about DS_HASHSIZE and DS_POOLSIZE which configure the distribution cache, not about DD_HASHSIZE and DD_HASHMAX which configure the dictionary cache and which use a mutex with a different name. Michael Colin Dawson wrote: > I have DD_HASHSIZE=997 and DD_HASHMAX=27, these numbers were arrived at by > running a script I downloaded from the s/w repository for calculating > DD_HASHSIZE and DD_HASHMAX values. I have just rerun it and for a HASHSIZE > value of 997 HASHMAX should be a minimum of 40. I'll check that when I next > get a chance to stop/start the server. > > > > > > > > >>-----Original Message----- >>From: Michael Mueller [mailto:m.mueller01.delete_me@kay-mueller.de] >>Sent: 13 March 2005 13:25 >>To: colin@colindawson.com >>Subject: Re: Latch Waits >> >>Colin, >> >>I assume you are running some 7.3x version? The mutex name you see is >>truncated and really is "global cache mutex". This is either the mutex >>protecting the distribution cache (holding update statistics medium and >>high >>information) or the one for the stored procedure cache. If you are able to >>catch a thread that is mutex waiting with "onstat -g stk <tid>" >>I should be able to tell which one it is. But maybe you can guess anyway. >> >>Maybe this is not related to your changes. You should try and set these >>onconfig parameters higher: >> >>for stored procedure cache: >> PC_POOLSIZE, PC_HASHSIZE >> onstat -g prc>> >>for distribution cache: >> DS_HASHSIZE, DS_POOLSIZE >> onstat -g dsc>> >>Michael >> >> >> >> >> >>Colin Dawson wrote: >> >>>Hi, >>> >>>Yesterday I started getting a lot of latch waits - onstat -u had "S" >>>in the 1st position in the flags column. onstat -g lmx showed a lot of >>>processes waiting on a mutex called "global cach" I cannot find much >>>information about this mutex. >>> >>>The question I'm asking is why would I get this situation? It has been >>>running fine previously. I did change BUFFERS from 1900000 to 4000000 >>>and RESIDENT from -1 to 0. So why would changing these two parameters >>>cause the mutex waits. When the change was made checkpoints went up >>>from 1-3 seconds every 15 minutes to 5+ seconds every 5 minutes >>>(leaving the interval at 15 minutes meant we had checkpoints > 10 >>>seconds) >>> >>>Some onconfig parameters: >>>LRUS 511>>>CLEANERS 127 (onstat -u shows some page cleaners not doing any work - >>>no of >>>writes=0) >>>CKPTINTVL 900 >>>LRU_MAX_DIRTY 1 >>>LRU_MIN_DIRTY 0 >>>RA_THRESHOLD 4 >>>NOAGE 1 >>>NETTYPE tlitcp,15,256,NET>>> >>>onstat -u shows approx 1300 connections>>> >>>Any insight into why I had the above problem would be appreciated. >>> >>> >>> >>>Many Thanks >>> >>> >>>Colin Dawson >>> >>>www.ladbrokes.com >>> >>> >>>sending to informix-list >> > > > sending to informix-list