Re: Logical Logs Size

Hi Mary, It's been a while since we talked. >Our experience has been with running small sized logical logs in our >continuous logging environment in the past (2k to 8k in size). At >approximately 4k size, our application will fill a log on the system >we are discussing in as little as 2 minutes and in as long as 30 minutes >given the varying times of the day and the load. > Is the log 2K or 2000? Since the size of the log files is in K, I'm guessing that you really have 2Meg log files. The main reason that sites have made small log files in the past is to try to get them to a backup media as soon as possible. This way they can recover a bit faster in the event of a disk head crash. I think that it's probably better to use mirrored disks for the log files for this purpose. At one point, I took this same approach. The size of the log files do not determine the size of the log buffer. By having larger log files, you do not take up any more memory. Also, you do not have to worry so much about a long transaction abort taking so much time because you can control that by using LTXHWM. There is a bit of overhead with switching log files, so I think that if you are switching every two minutes, that is a bit excessive. Generally I suggest that people size their systems so that a log switch occurs about one every 15 min. unless they are using Enterprise Replication (in which case I suggest once an hour). So - basically I agree with your admin group - probably make them about 10 times larger than they currently are. Now about those checkpoints --- I really think that 150 seconds apart are a bit too often. But if your user's are not complaining - I'm not either. Your physical log file does not have a direct relationship with the logical log file. However the physical log file does have a directly relationship with checkpoint frequency. If the physical log file is too small, then your checkpoints will actually occur before you reach the configured checkpoint interval. >The jutebox can't keep up with the logs filling in minutes, the >storage media is meant for less often backups, but to stream a lot of >information quickly to the tapes. > >The primary question, HOW BIG TO MAKE THE LOGICAL LOGS? Our system >administration group would like to see 50 megabyte logs. That's way >too much shared memory for me, or is it?? > >I'm looking for experience and recommendations on suggested "larger" >sized logs. Please also comment on the following concerns if you can: > > - shared memory (allocating to a bundle to large logs?) > - long transactions (increasing the log sizes, have you decreased > the long transaction high water marks? We're > at 50/60 for LTXHWM/LTXEHWM) > - recovery (as long as checkpoints are in line, are you > safe? I've been in recovery on a long > transaction and hence my concern above about > rolling back forever and a day on a long txn) Madison Pruet