Re: Redo logs - striped vs stand alone drive

Separating the redo logs from the datafile is a good practice for distributing I/O and gaining improvement on performance . It is also good since it may increase the probability of recovery incase there is a database crash and you loose a disk .... On this case you don't loose a datafile and a redo log at the same time .

About not having the redo log placed on striped disk ..... I didn't understand the wisdom about this until recently ....The nature of redo log is 100 sequential write , and raid stripping performs best on random reads . However if you are sure that you have adequate disk buffering , placing the redo log on stripped disk shouldn't be much of a problem ...Putting the redo log on a separate non stripped disk is the probably a good option to consider ....

joseph_sumalbag_at_bose.com

x wrote:

> According to: http://www.oramag.com/archives/15DBA.html
> Redo log files store the records for each transaction in the database. The
> Log Writer (LGWR) background process writes transaction entries to the redo
> log files. At the same time, the Database Writer (DBWR) background process
> concurrently writes the data in the transactions to several tablespaces
> (such as the RBS rollback segments tablespace and the DATA tablespace).
> Thus, even though the datafile I/O may be properly distributed, there will
> be contention between the DBWR and LGWR background processes if a datafile
> is stored on the same disk as a redo log file.
>
> And: http://www.oramag.com/archives/55OPOPEN.html
> Redo log access is 100-percent sequential I/O and needs to be isolated if
> possible. Speed of the redo log is essential to system performance. If
> possible, these drives should be fault tolerant. Fault-tolerant hardware
> provides maximum performance and reliability.
> Redo log archive files are also 100-percent sequential I/O and should be
> isolated for maximum performance.
>
> I have also heard that striping is not ideal for redo logs because they
> perform best when the next write position is the next block on the same
> physical disk. Is striping less ideal or is it just that if striping, the
> redo log should be the only file on that striped set? What kind of
> performance gain would be expected from redo logs on its own disk vs. shared
> on a striped set with system and temp tables?
>
> xenophon_at_usa.net
Received on Fri Apr 03 1998 - 00:00:00 CST