Re: Informix performance Issues
Tim Chunk writes are more efficient, but they tend to freeze the system during checkpoint giving users the impression that the system is "hung". It is better IMHO to have LRU writes occuring between checkpoints where the overall user-experience is a smoothly running system. AFAIK, this is primarily a consideration in OLTP systems where response times are expected to be small. Probably in the DW world, it may be better to have chunk writes since all the inserts occur in batch. Just my $0.02 (I know this question was addressed to Art, and perhaps he would do a much better job of answering it :-)) Sujit Tim Schaefer <tschaefe@mindspring.com> on 06/10/99 08:31:46 PM Please respond to Tim Schaefer <tschaefe@mindspring.com> To: informix-list@iiug.org cc: (bcc: Sujit Pal) Subject: Re: Informix performance Issues "Art S. Kagel" wrote: > > Fg Writes LRU Writes Chunk Writes > > 0 0 27085 > > You show 100% chunk writes. No matter what the manual states chunk > writes are death! You MUST detune the LRU_MIN/MAX_DIRTY from the 90/80 > to a more reasonable 50/10. You do not need the 2/0 that a heavy > transaction system needs but I'd bet your checkpoints are taking over > a minute out of every 50 minutes when they could be taking 10-20 > seconds instead. > <...> > > Art S. Kagel Art, Not only do the manuals state chunk writes are best, but the training states this as well. Could you elaborate on why chunk writes are bad? This is counter to not only the docs but the training. Thanks, Tim -- - -- --- Tim Schaefer ---- tschaefe@mindspring.com --- http://www.inxutil.com -- -
Thanks, ( and this is for Art's comments too ) I've heard so many differing concepts of what is right and wrong, but tend to at least try and stay away from the Fg Writes. I've been told to crank down LRU's for OLTP, loosen them up for DSS, it seems to be staying this way for now. Your comments are appreciated. Tim Sujit.Pal@BankAmerica.com wrote: > > Tim > > Chunk writes are more efficient, but they tend to freeze the system during > checkpoint giving users the impression that the system is "hung". It is > better IMHO to have LRU writes occuring between checkpoints where the > overall user-experience is a smoothly running system. AFAIK, this is > primarily a consideration in OLTP systems where response times are expected > to be small. Probably in the DW world, it may be better to have chunk > writes since all the inserts occur in batch. > > Just my $0.02 (I know this question was addressed to Art, and perhaps he > would do a much better job of answering it :-)) > > Sujit > > Tim Schaefer <tschaefe@mindspring.com> on 06/10/99 08:31:46 PM > > Please respond to Tim Schaefer <tschaefe@mindspring.com> > > To: informix-list@iiug.org > cc: (bcc: Sujit Pal) > Subject: Re: Informix performance Issues > > "Art S. Kagel" wrote: > > > > Fg Writes LRU Writes Chunk Writes > > > 0 0 27085 > > > > You show 100% chunk writes. No matter what the manual states chunk > > writes are death! You MUST detune the LRU_MIN/MAX_DIRTY from the 90/80 > > to a more reasonable 50/10. You do not need the 2/0 that a heavy > > transaction system needs but I'd bet your checkpoints are taking over > > a minute out of every 50 minutes when they could be taking 10-20 > > seconds instead. > > > <...> > > > > Art S. Kagel > > Art, > > Not only do the manuals state chunk writes are best, but the training > states this as well. Could you elaborate on why chunk writes are bad? > This is counter to not only the docs but the training. > > Thanks, > > Tim > -- > - > -- > --- Tim Schaefer > ---- tschaefe@mindspring.com > --- http://www.inxutil.com > -- > - -- - -- --- Tim Schaefer ---- tschaefe@mindspring.com --- http://www.inxutil.com -- -