Re: buff write cache to low

Jonathan Leffler wrote: > > >From: David Henseler <henslerd@bscc.bls.com> > >Date: Mon, 24 Feb 1997 13:45:57 -0500 > >X-Informix-List-Id: <news.34348> > > > >I'm in the process of performance tuning an application running and I'm > >finding the buff write cache falling well below 50% while the read cache > >stays above 97%. The specifics of the application are three main > >applications each one horizontally partitioned ( multiple iterations of > >the same application running different subsets of data feeds). > > These applications sound as if they read the database more than they write > to it. > > >The box is a 16 way SGI with 2gb of mem and a 4 shelve MTI disk array. > >The meaningful config parms are as follows. > > > >BUFFERS 300000 # dlh020597 > >[...] > >LRU_MAX_DIRTY 6 # dlh020597 > >LRU_MIN_DIRTY 3 # dlh020597> > I've trimmed down the meaningful parameters to the 3 above. I suspect that > you have a read-mainly database; most operations on the database are read > operations, and the ratio of disk reads to disk writes is more than 100:1. > Your LRU_{MIN|MAX}_DIRTY parameters also suggest that you have found that > you don't fill up much disk space between checkpoints. > > Under these circumstances, IM(not so very)HO, the write cache ratio doesn't > matter in the slightest; there isn't enough write activity for there to be > much room for improvement in performance, even if you managed to get the > write cache ratio magically to 100%. > > On big systems, the read cache is critical, and the write cache is often > (but not always) non-critical, especially in applications which are not > intensively OLTP. > > Yours, > Jonathan Leffler (johnl@informix.com) #include <disclaimer.h> Hi, I fully agree to Jonathan's words. As long as you don't have I/O waits ( output waits ), the write cache will be non-critical. Bye Stefan PS: I didn't find the size of the physical log file. It's interesting just to find out, if a checkpoint takes place very often.