Re: logical logs

On Thu, 11 Dec 1997, Thuc Nguyen wrote: > Hi > > I am running Online 7.23 on HP 10.20 > > Why was logical log backed up every 2-3 minutes? > Is there any hint not to make that backup too fast? > Here is the info... > > # Physical Log Configuration > > PHYSDBS phydbs # Location (dbspace) of physical log > > PHYSFILE 60000 # Physical log file size (Kbytes)> > # Logical Log Configuration > > LOGFILES 20 # Number of logical log files > LOGSIZE 6000 # Logical log size (Kbytes) Sure, your log files are too small for the level of update activity that you apparently have. Make each log file 60000K->100000K and you will change logs every 10-20 minutes during peak activity instead of the current 1-2 minutes. BTW: I noticed your NETTYPE parameter for onipcshm in CPU VPs with 1 listener. If you increase this to 4 listeners there will be a listener in each CPU VP and you will 1) spread the load across the CPU VPs better and 2) be more responsive. The problem, I have found, is that the VPs with listeners take most of the work onto themselves which makes them too busy to listen but when they free up to listen they are not busy so they add the next job to their own queue instead of handing it off to another CPU VP. So 3 of your CPU VPs are in wait loops waiting for work most of the time while CPU VP1 is swampped. Check the CPU time distribution over the VPs in onstat -g glo with the current config and with 4 listeners and you can convince yourself. We started with 28 CPU VPs and 8 listeners as Informix consultants thought we should only need a few listeners. There were always some applications that were periodically paused waiting for service but 20 of the CPU VPs were never busy. We increased to 28 listeners and voila! No more delays and better VP utilization. Happy databasing. Art S. Kagel, kagel@bloomberg.com