One big chunk or many smaller ones?

IDS 10.0 on RHEL AS 4: We're setting up a new database server for an OLTP application (24x7 website). We've agreed on 100Gytes of application dbspace in a single dbspace. What are poeple's opinions on having 10 x 10g chunkis rather than 1 x 100g one? On an issue that may or may not be related, we were having performance issues (still are in fact) on a similar environmment, albeit 10.0FC3 on Solaris 9, where the symptom was unexpectedly long checkppoints (up to 10s, and slow flush times as we observed them). IBM UK Tech Support advised us to split the data across multiple dbspaces to try to speed up checkpoint activity. We did this: it made no noticeable difference to checkpoint speeds but is more work to administer. I'd be interested in the theories and practial experience of others, particularly where we're having a single dbspace. Thanks -- Neil Truby t:01932 724027 Director m:07798 811708 Ardenta Limited e:neil.truby@ardenta.com

Neil Truby wrote: > IDS 10.0 on RHEL AS 4: > > We're setting up a new database server for an OLTP application (24x7 > website). We've agreed on 100Gytes of application dbspace in a single > dbspace. What are poeple's opinions on having 10 x 10g chunkis rather than > 1 x 100g one? > > On an issue that may or may not be related, we were having performance > issues (still are in fact) on a similar environmment, albeit 10.0FC3 on > Solaris 9, where the symptom was unexpectedly long checkppoints (up to 10s, > and slow flush times as we observed them). IBM UK Tech Support advised us > to split the data across multiple dbspaces to try to speed up checkpoint > activity. We did this: it made no noticeable difference to checkpoint > speeds but is more work to administer. > You get one page cleaner allocated per chunk at checkpoint time. If you even up with 1 page cleaner doing all the work checkpoint times fly up. Make sure you have CLEANERS set high enough (at least 48, perhaps even up to 127) and use enough chunks to get the parrallelism you need! Run onstat -z and onstat -u to see how the checkpoint I/O is spread across the page cleaners. > I'd be interested in the theories and practial experience of others, > particularly where we're having a single dbspace. > > Thanks > -- > Neil Truby t:01932 724027 > Director m:07798 811708 > Ardenta Limited e:neil.truby@ardenta.com