Re: reasonable checkpoint interval

In article <4ra0km$bi2@sjx-ixn3.ix.netcom.com>, Stuart Cracraft <cracraft@ix.netcom.com> writes >What is considered a reasonable informix checkpoint interval? > >I'm currently investigating a problem on a system with the checkpoint, >as shown in /usr/informix/online.log, occuring every 6 minutes or so. >This is on a Unix box with Informix 7.12.UC1. > >Users on PC's have a GUI application to do queries into this Informix >Online SQL engine. Every few minutes, their PC's freezeup (all >simultaneously) for up to 5 minutes reducing their productivity by >orders of magnitude. > >What checkpoint intervals are considered reasonable? Is there >such a thing as having checkpoint intervals too close together? >What are the Informix defaults? > >Users of this system are doing a moderate number of queries typically >against extremely large tables all indexed and multiply so. Meanwhile >a large number of system batch jobs are running doing lots of table >queries. > >Thanks ahead. > >--Stuart > > Normally a checkpoint interval of 5 minutes (the default) is OK. You need to either a) Reduce LRU_MIN_DIRTY and LRU_MAX_DIRTY from the defaults of 50/60 to 30/40 to minimise disk writes at checkpoint time OR b) Increase the number of page cleaners (CLEANERS) to the number of disks across which you OnLine Instance is spread. OR c) Increase LRU queues to the value in b). OR b) Use tbstat -d to look at the chunk ids for your OnLine instance The chunks should be spread across disks in chunk id order i.e. chunk 1 on disk 1, chunk 2 on disk 2 etc. This is because OnLine allocates page cleaners to clean chunk in chunk id order. Unfortunately to fix this you will have to configure your Informix Instance from scratch as chunk ids are assigned when you create the chunk. This can make quite a BUG difference to checkpoint times... ANy chance of adding this to the FAQ, Kerry? -- David Williams