Re: long checkpoint duration

Floyd Wellershaus wrote: > Hi, > The age old question I guess. But our checkpoint durations are > consistently near 20seconds. > > ids10.0 fc5 on ibm p570 running aix 5.3 7 cpus 41g ram. > ckptintvl is 300, I am considering halving that. > We have 128 lru's and 128 cleaners > Art's ratio tool shows all good except a high btr. > We have 2557440 buffers, but I am pretty sure if we add much more > we'll start paging. Anyway , that probably isn't affecting the checkpoint. > > onstat -g iov shows the inverted pyramid with 7 kio threads.> > Aside from worrying about fragmentation and data distribution issues > in our FastT900 san, is there anything I'm missing that I should look > at real quick to reduce the chkpoint interval ? LRU_MIN_DIRTY and LRU_MAX_DIRTY (or the equivalent BUFFERPOOL option) settings control the number of dirty buffers that need to be flushed at checkpoint time which, aside from user threads holding critical section locks, is the most common cause of long checkpoints. Try reducing those settings. In IDS 10.00 you can set these to fractional values to which is usually neccessary for larger buffer caches with FUZZY checkpointing turned off. Take the number of dirty buffers flushed at checkpoint time and the desired checkpoint length (say under 2 seconds) and compare the later to the current checkpoint time. Apply the factor to the current LRU_MIN/MAX_DIRTY parameters to reduce the number of dirty buffers to the number your system can flush in 2 seconds. Art S. Kagel Oninit > > Thanks, > Floyd > > > > ------------------------------------------------------------------------ > > _______________________________________________ > Informix-list mailing list > Informix-list@iiug.org > http://www.iiug.org/mailman/listinfo/informix-list > =========================================================================================== Please access the attached hyperlink for an important electronic communications disclaimer: http://www.oninit.com/home/disclaimer.php ===========================================================================================