Re: Better Manuals (was: Re: Mirroring)

On Mon, 28 Jul 1997, Clem Akins wrote: > Hi Art, > > >I know what the manuals say (one cleaner per disk and fewer LRUs is > >better). The manuals are WRONG. The Informix benchmark people long ^^^^^^^^^^^^^^^^^^^^^^^^^ This is not new information, just not well disseminated. I would hope that the people in the benchmark group would have spoken to the documentation group long ago. I got the information from Alan Slutsky, who left Informix last year, over six years ago when he was doing a lot of competitive benchmarks. > >ago proved that and I have verified it myself. If your goal is to best > Have you sent this observation to the doc@informix.com alias? They > invite your opinion on the last page of the Introduction in each book, and > appreciate suggestions on how to make the manuals better. I recommend > that you write up a page showing how you think the manuals *should* look > and send it on. The problem is that the manual is not exactly wrong, that is actually a little strong. The recommendation in the manual was developed with the help of the programmers responsible for the page cleaner/checkpoint code and reflects the best recommendation if your goal is to play nice and not have the Informix engine take over the machine. However, the manuals do not discuss what to do if you are building a dedicated server machine or one that only runs the Informix engine and Informix database applications. This is partly moot with the advent of 7.xx as the waters are much muddier. Add to the I/O performance tunables #Buffers, #LRUs, #Cleaners, AIO VPs -vs- KAIO (different preferences depending on machine and #CPUs), #AIO VPs (or #CPU VPs if KAIO), larger disk farms (5.0x only supported a practical maximum of about 54 chunks, 7.xx can have 1024). What would I write up? Hey how about a book for Informix Press detailing Administration and tuning tips? I could do that! Art S. Kagel, kagel@bloomberg.com