Re: PHISICAL LOGS configuration

Im Artikel <59m9tq$6gh@cssun.mathcs.emory.edu>, mpisano@ed.gba.gov.ar ("Mauro G. Pisano") schreibt: > Here are my questions : > > 1. Is the disk space well managed ? > 2. wich is the On-line's space, is all the space or only the root >dbspace's space ? > 3.shoul I prefer few and large logical logs or many and small ones ? > >Thanks very match in advance for your help. > > Here are my steps to configure a 5.0 database: First, I create at least 3 raw devices: - one for rootdbs (approx. 150MB) on disk 1 - one for the logical logs (approx. 50MB) on disk2 - one or more for the data dbspace ond disk 3, 4, 5 Note: - never create any dbspace on the first disk (disk 0). This is a high frequent disk under UNIX. - do not use the disk on which TMPDIR (e.g. /var/tmp) points. This place will be used from INFORMIX for sorts, etc. - do not use any disk, where a swap space is - for the data dbspaces it is better to use virtual disks (max. 2GB) Then, I create the logical logs - logsize is 1000KB (nice sice :-)) - number of logs on low frequent databases (low users): 12 - number of logs on high frequent databases (many users): 24 - physical log size = 1/3 logical log size Last startup tuning: - Set Checkpoint Intervall to 10 Minutes - Create one page cleaner for each physical/virtual disk WARNING: If you have more than 1 partition on your physikal disk and you have more than 1 disk, than you have to watch for the order you add chunks to a dbspace. Cause INFORMIX runs page cleaners as follows: 1st page cleaner for the primary chunk, 2nd page cleaner for the 2nd chunk and so on. But you can use only one page cleaner per physical/virtual disk. If the 1st and the 2nd chunk on the same physical/virtual disk, is the result as follows: The 1st page cleaner works, the 2nd sleeps. - Try to use buffered logging. This log mode is a little bit dangerously than unbuffered logging, but you will get much more performance. DONT FORGET THE LTXHWM/LTXEHWM parameters set to 60/70. If you do not set these parametes, in cause of a long transaction rollback, your database will be damaged and never can be recovered! OK, this were my startup tunigns. After that, you have to watch for checkpoint intervalls - informix.log long transaktion rollbacks checkpoints wich runs longer then half a second (This is a personaly breakpoint) if checkpoint intervall occurs less than 10 Minutes, try to resize the physical log. (Dont't forget the logical logs 1:3) if you can find many "long transaktion rollbacks" then you have to add more logical logs. if the application stops, during a checkpoint, you have to check your page cleaner or resize your buffers. I hope, that helps. Greetings Volker Frankle