Hi,
reading the messages from this list, sometimes I have doubts about the
relation between LRU, CLEANERS, AIOVPs, BUFFERS and NUMCPU.
I think that LRU and CLEANERS must be identical values. But there exists any
predefined relation between the others values ? In my environment, a Sun
E-6500 with 14 processors and 9GB RAM, I have 800.000 buffers, 16 LRU, 16
CLEANERS, 100 AIOVP, 13 NUMCPU
Thanks in advance
fred
"Art S. Kagel" wrote:
> Bridget Reitsma wrote:
> >
> > Hi
> >
> > I'm using informix 730UC5 on a Sequent box, I thought I had everything
> > set just right, that was till Wednesday this week, when my checkpoints
> > started rocketing. Normally my checkpoints average at about 4 seconds,
> > now they are averaging at about 60 seconds.
>
> > My LRU MAX is set to 2 my LRU_MIN is set to 0. The have 17 LRU queues
> > and 8 CLEANERS. The checkpoint interval was recently changed to 600
> > seconds - I have have changed this back to 300 seconds. (I have
> > attached a copy of my onconfig file.)
>
> It might be the increase in the checkpoint interval, but I doubt it. BTW
> increasing CLEANERS to 17 or increasing both LRUS and CLEANERS to 31 or 32
> MAY reduce the checkpoint duration but I don't think that any of these
> items is to blame for such a substantial increase. More likely it is
> a process which is in a critical section causing the checkpoint to wait
> for it to release its latch before beginning. The checkpoint blocks new
> requests for a critical section latch and then waits for the existing
> latches to clear before continuing with the checkpoint. So if there are
> a later number of processed in crcitical sections, especially if they are
> waiting for each other to free resources, this can cause the checkpoint
> to become blocked.
>
> Just now I had a server checkpoint for 15 seconds with only 53 dirty
> buffers and
> 3 pages of physical log and 3 pages of logical log buffer to
> flush to disk. I run 127 LRUs and 127 CLEANERS and have 49 AIO VPs. This
> checkpoint should have taken less than a second. It did not because I
> have 21 VERY active middleware processes querying the system.
>
> > There hasn't been a major increase in users the load is only slightly
> > higher than normal. It is the end of the month after all. But does
> > anyone have any ideas of what other step I could take to decrease the
> > checkpoint interval?
> [SNIP]
>
> Art S. Kagel