RE: Reducing checkpoints

-----Original Message----- From: informix-list-bounces@iiug.org [mailto:informix-list-bounces@iiug.org] On Behalf Of TBP Sent: Tuesday, June 20, 2006 3:13 PM To: informix-list@iiug.org Subject: Re: Reducing checkpoints Superboer wrote: > hope NO RAID 5 NO RAID 5 NO RAID 5 > > for starters: > > CKPTINTVL 360 # Check point interval (in sec)> > i would set it much bigger so you get a checkpoint each half hour > be aware that fast recovery may take longer in case of a crash. How would this help reduce the checkpoint *duration*? I would suggest that we have some form of O/S or H/W issue here. The instance is not configured that big : BUFFERS 400Mb. LRU_MAX_DIRTY 1 - which represents about ... 4 Mb - not a lot really,and certainly not enough to cause a 58 second checkpoint. CLEANERS 127 LRUS 127 But we don't know how many chunks, nor whether KAIO is in use - presumably not KAIO 'cos RESIDENT is not set. NUMAIOVPS 8. But, to me at least, looks like something outside of the engine. _______________________________________________ Informix-list mailing list Informix-list@iiug.org http://www.iiug.org/mailman/listinfo/informix-list ----------------------------------------- -- this email delivered by hermoth Could be a thread taking a long time to complete a critical section for whatever reason, long checkpoints aren't always a result of lots of dirty buffers. Try a couple of `onstat -u | grep X` during a long checkpoint, this will show you if a checkpoint is waiting for a thread to exit a critical section. Also I have seen a lot of btree cleaner requests cause silly long checkpoints, what does an onstat -C return? I agree with upping LRU MIN/MAX to 1/2, but this prob won't reduce checkpoint duration. What do you see during a checkpoint? Does onstat -R show a large number of dirty buffers and it is just taking a long time to write them to disk or is the number of dirty buffers constant (i.e. no chunk writes are happening and the checkpoint is waiting on something else)? If you are spending all of that time in the chkpt doing chunk writes, you can experiment with onmode -B to do a nonblocking dirty page flush before a checkpoint. Check out this onmode -B link http://groups.google.com/group/comp.databases.informix/browse_frm/thread /f8ce18266ee2fddd/12eebaedc22da9e0?q=%22onmode+-B%22&rnum=1#12eebaedc22d a9e0