multiplexed redo logs - how enhance performance

[ggnanaraj]

We know that we need to multiplex redo logs, so we have more than one redo log member in each group on seperate physical disks to safeguard against media loss.

Now, how does this multiplexing of redo logs enhance performance. In other words, can we enhance the performance of 'multiplexing of redo logs'.

TIA.

[DaPi]

Since a commit is not complete until the redo information is physically written to all members of the group, you have actually increased the chances of something slowing down the commit! (Ditto for log switches). You need to make sure that all the redo disks have no other activity that will conflict with writing the redo logs.

[ggnanaraj]

I have collected some more information about this...

Asynchronous I/O will be used, if available to make the write parallel, otherwise these writes are done sequentially one member after the other. However, LGWR has to wait until I/Os to all member logfiles are complete before the 'log file parallel write' wait event is complete.

BTW, to reduce 'logfile parallel write' wait time, the following can be done...
- reduce redo generated (NOLOGGING)
- reduce redo log members (RAID 1 - with 1 member per group?)
OR
- use RAW devices
- place members on storage locations to avoid parallel write contention
- use faster disks

[SANJAY_G]

Originally posted by ggnanaraj
Now, how does this multiplexing of redo logs enhance performance.

BTW, who said that??

[grjohnson]

Originally posted by ggnanaraj

BTW, to reduce 'logfile parallel write' wait time, the following can be done...
- reduce redo generated (NOLOGGING)
- reduce redo log members (RAID 1 - with 1 member per group?)

And neither are satisfactory options.

[tamilselvan]

By multiplexing redo logs, archive process can read any one of the redo logs that has been finished first. This perf gain is very small. By multiplexing redo, you are safe guarding the redo log.

If you really want to enhance the perf of redo, then you have to place them in RAW DEVICES.

If you put them in FILE SYSTEM, then there is a small lock conflict occurred becuase of WRITER (LGWR) blocks READER (ARCH) and the internal structure of inodes handled by UNIX OS. This lock conflict will not occur on RAW DEVICES.

Tamil