Hi, Jon and Mike, My understanding is that Oracle always performs single write to headers of control files, log files and datafiles, and parallel write to the body of control files, log files and datafiles. The wait events are correspondingly control file / log file / db file single write, control file / log file / db file parallel write. However, if the OS does not support asynchronous I/O, parallel write will not be performed and serial write is used instead. I vaguely remember only AIX among all UNIXes has problems with AIO, so it usually is not a problem. In any case, Steve Adams' article is good reading on this topic: *****Begin copy (www.ixora.com.au/tips/creation/raw_log_files.htm) If there are multiple log file members, LGWR must write the redo data to each active member. If non-blocking (asynchronous) writes are possible, and if the disk_asynch_io parameter is set to TRUE, then LGWR will perform these writes in parallel. However, if asynchronous writes are not available, then LGWR must write to each active log file member in series. This effectively doubles the response time of redo writes. Do not be deceived by the fact that the wait for log file writes is called a log file parallel write wait. If LGWR cannot write in parallel, it won't. LGWR can only write in parallel if asynchronous writes are available ******End copy Regardless AIO, striping redo log files increases LGWR performance. As Jon explains, this hardware level trick is not known to Oracle and speeds up writes behind the scenes. RAID-5 usually is not a problem because what vendor is still selling RAID-5 without a big cache nowadays? It *is* still a problem if you're writing on a massive scale such as in database backup. Yong Huang yong321 AT yahoo.com Jon Crisler wrote: > Hmmm, interesting. I found some specific docs that advocate redo on striped > volumes. The goal was to increase the rate of I/O's that can be supported. > > I tried moving redo from a single disk to a striped disk (2 disks) and my > average I/O wait went from 200ms to 30ms, measured over 2 days. So it > actually (greatly) helped in my case. > I realize that redo is sequential, but how would striping (raid 0+1 in my > case) hurt ? I do know that RAID-5 is to be avoided since most I/O (or at > least 50%) is writes. In any case, the striping is transparent to Oracle, > so serial vs. parallel should not be an issue. > > Mike, could you expand on this issue ? > > -----Original Message----- > From: Michael Ault [mailto:mikerault AT yahoo.com] > > Actually Oracle says redo should not go on striped > volues since redo logs are written in serial and not > in parallel. You can go with a raid 1 but a raid 0 or > raid 01 or 10 is not suggested. RAW would be faster. > > Mike