The average rotational latency for a full stripe write to a stripe set in
d/(d+1) of a full rotation where d is the number of disks in the stripe. That
reduces to 1/2 of a full rotation for just one disk, but 2/3rds or more for a
stripe. Therefore, to get any benefit whatsoever from striping log files, the
saving in transfer time must exceed the extra 1/6th of rotation wasted on
rotational latency. Therefore, the average log write size must exceed 1/3rd of
the track length of the disks, or about 24K for a typical disk.

There are two sorts of redo intensive systems. Those that commit frequently, and
those that don't. If your system commits frequently, then your average log write
size will not exceed 24K and striping the log files would actually hurt
performance no matter how intensive the redo generation is. You can determine
the average write size using the lgwr_stats.sql script from the Ixora web site.
If on the other hand you have a system that does large scale data changes and
seldom commits, then there is a possibility that striping the log files would be
beneficial, but probably much less so than you would have expected.


-----Original Message-----
Sent: Saturday, 26 August 2000 6:58


Suppose we want to increase the bandwidth for the log file disks, and based on
the redo write rate we find that 3 disks are required.  On what basis would we
set the stripe size so that log writer writes to all these disks in parallel.
Assumption being that the log files are on raw partitions.