Re: Slow inserts

Art S. Kagel wrote: > Ben Thompson wrote: > > Ben, do you perform large numbers of deletes on this table? The > periodic large numbers of IOs on the table's chunk may be BTREE > cleaner/scanner activity. BTW Version and platform info would be > helpful, especially since the BTREE cleaning functionality has changed 3 > or 4 times over recent releases to improve things. Contrary to what I posted before, it appears that there are deletes being performed on this table - a few thousand rows only though. I have actually managed to reproduce this on a test box which is part way to getting a solution. The symptoms are that if I set the BUFFERS parameter on the test system quite low, and do deletes, the whole system runs a bit more slowly. The effects of slow running continue for some time after the deletes have finished and even if you remove all connections to the database server, there are still a lot of reads going on on this table. Getting rid of the primary key and performing the same tests improves things somewhat. Ben.

Ben Thompson wrote: > Art S. Kagel wrote: > >> Ben Thompson wrote: >> >> Ben, do you perform large numbers of deletes on this table? The >> periodic large numbers of IOs on the table's chunk may be BTREE >> cleaner/scanner activity. BTW Version and platform info would be >> helpful, especially since the BTREE cleaning functionality has changed >> 3 or 4 times over recent releases to improve things. > > > Contrary to what I posted before, it appears that there are deletes > being performed on this table - a few thousand rows only though. I have > actually managed to reproduce this on a test box which is part way to > getting a solution. The symptoms are that if I set the BUFFERS parameter > on the test system quite low, and do deletes, the whole system runs a > bit more slowly. The effects of slow running continue for some time > after the deletes have finished and even if you remove all connections > to the database server, there are still a lot of reads going on on this > table. Getting rid of the primary key and performing the same tests > improves things somewhat. To add more information, the BTREE scanner is going like mad when this occurs. I didn't know the "onstat" for this one but Informix Server Administrator can be very handy... So it looks like Art was right. Ben.