Re: Online Checkpoint Issues

In article <1992Sep3.124413.9713@xact.demon.co.uk> ptm@xact.demon.co.uk (Paul Thomas Mahoney) writes: >We run Online with about 200 users and I have some questions that I would >appreciate some comments to. The configuration is as follows: > > - The OS is SCO 3.2.2 > - We have 9 client machines (Dell 33MHz 486) each supporting 25 users > running our application, written using ESQL/C. > - These clients call across the net to the server (Dell 50MHz 486 with > about 2GByte disk). > >Thus about 200 connections over I-Star to the engine. All databases use >unbuffered logging and there is a 525MByte tape to do continuous backup. >Overnight a level 0 backup is done to DAT. Shared memory is about 14MByte. >Checkpoint interval is 3 mins. > >Now the problem: > >Each checkpoint is taking 30-45 seconds to complete with about 3 megs of >log to be written to disk. As the checkpoint starts, all currently running >processes freeze. Then the disk activity takes place for 15-20 seconds and >then the user tasks wake up, some much more slowly that others. > >Looking at the priorities of the sqlturbo processes on the server during >checkpoints we find that, during normal running, they are in the mid 20's >but by the end of the checkpoint their priorities have dropped (risen?) to >65-70. I suppose this accounts for their slowness in re-starting. Try setting your LRU_MAX_DIRTY and LRU_MINDIRTY tbconfig params lower, something like 15 and 5 or 10 and 5. This should cause more background i/o by the page cleaners so there isn't as much to do at a checkpoint. mark