Re: About RA_PAGES

"Dpto. Informatica" wrote: > > mars1972@my-deja.com wrote: > > > > > RA_PAGES 1000 # Number of pages to attempt to read> > > > ahead > > > > RA_THRESHOLD 500 # Number of pages left before next> > group > > > > > > Or the bigger question (to me): > > > How did you arrive at those values for your RA variables?? > > > > Definitely. In the Performance Tuning guide, they state quite clearly > > that the RA_PAGES shouldn't be set higher than 32, for this very reason. > > Try RA_PAGES at 32 and RA_THRESHOLD at 16 or 24 so so. What's > > happening is each of the aoi threads is being forced to read 1000 * > > PAGE_SIZE, or probably either 2 or 4 MB into memory for every 500 read. > > I bet your ra pages used is in the gutter. Well 32 is based on a single IO. In DSS environments you might go bigger, but at the expense of more IO. > Sorry, I haven't the documentation 'Performance Tuning Guide', I've > only > the 'Performance Guide', where I've used the follow formula, in page 3-60: That's the right manual. > RA_PAGES = (BUFFERS * bp_fact) / ( 2 * large_queries) + 2 > RA_THRESHOLD = (BUFFERS * bp_fact) / ( 2 * large_queries) - 2 > > bp_fact = Portion of data buffers to use large scans that require > read-ahead > i.e.: 75 percent off buffers. > > large_queries = number of concurrent queries that require read-ahead that > you intend to support. > > This information is OLD or is more new than you have ?. No, this is current. But what values did you use in your calculation? You should end up with two numbers that differ by 4, not 500! > If my information is not correct, then I will use RA_PAGES = 32 and > RA_THRESHOLD = 24. That sounds good. Cheers, -- Mark. +----------------------------------------------------------+-----------+ | Mark D. Stock mailto:mdstock@mydas.freeserve.co.uk |//////// /| | http://www.informix.com http://www.informixhandbook.com |///// / //| | http://www.iiug.org +-----------------------------------+//// / ///| | |What year 2000 bug? year 2000 bug? |/// / ////| | |year 2000 bug? year 2000 bug? year |// / /////| | |2000 bug? year 2000 bug? year 1900 |/ ////////| +----------------------+-----------------------------------+-----------+