Re: Another checkpoint situation

At checkpoint time one page cleaner is assigned to queue up the addresses of modified buffers to the global file descriptors for its assigned chunk. (This assumes that you are not using Kernel IO) It appears that all of your work between the check- point was for one chunk and thats why the checkpoint duration is so long relative to the small amount of work performed (this causes all of your I/O to be single threaded). Use fragmentation to cause the I/O to be balanced across multiple dbspaces and therefore chunks - causing multiple parallel page cleaner processing. As a general rule: one page cleaner per chunk, one LRU queue pair for every 500 to 750 buffers and one AIO VP per global file desciptor (gfd from onstat -g ioq).