Re: Long checkpoints with no obvious cause

Tony Flaherty wrote: > Replying to my own post here but... > > It appears that when my checkpoints occur the physical log is practically > empty. Would I see an improvement if I actually increased the time between > checkpoints? This would I assume result in an increase in LRU writes > leaving less work for the checkpoint to do? > Nope. In fact, your onstat -F output indicatesLRU writes are less than 5% of total writes. This means that your dirty buffers are rarely reaching the LRU_MAX limit. Increasing the time between checkpoints could ensure that they will be closer to LRU_MAX - which could mean that there's more work to do at checkpoint. You probably want to reduce LRU_MAX and MIN to force more LRU writes. However, that still does not explain the spike in your checkpoint duration (which is the essence of your problem) I would pull up glance (since you are on HP) and monitor cpu and disk activity during a rogue checkpoint. Maybe there are other hogger processes using the cpu or the disk at the time. Or it could be related to a bug in your version. Rudy