Re: Log files on RAID 5...

"Kelly-Anne Frendo" <kfrendo_at_bvb.gi> wrote in message news:A3C801763DAFD411A57900508B9AAF972930_at_HEARTOFGOLD...
>
> Thanks for your mails, but can I just explain why I am a little
 confused...
>
> Okay, so anything sequential like online redo logs should not go
 on RAID 5,
> but it is okay for the archived logs isn't it? Because don't
 archive logs
> get written in one go when the log switch occurs? Or does even
 this writing
> to archive log file count as sequential?
>
> Kel.
>

Might I add to the questioning,

Log writer (LGWR) does a sequential "style" of writing, meaning it writes to the first log file until it's filled, then to the next, and so on, until the last one is filled, then it starts over with the first log file again (circular reuse). During this, are the redo entries written to the log files only contigously and in one big write? This writing is done at the file-system level (not device I/O or Raid system level). I assume that Archiver works very similar though copying whole file in one go.

Where is the problem with that and Raid 5 and how a Raid controller handles the actual disk write operations (I'm sure some implementations are better than other)?

I mean, generally Raid 5 has n+1 disks for a striped array with the parity evenly distributed. For every write there are data stripes writes and a parity caclulation and a write of parity to one of the disks (done in parallell?). So, even if the whole array width in stripes is used there is always a "penalty" for writes compared to reads from the array. If there is a high write to read ratio, at the disk level random or not, you "lose" performance. Since redo log files are almost never read from, then you should always avoid Raid 5? (That is if you need the performance, or have superflous cash supply).

Basically what I want to say is, I think there are many factors (many more than just that Lgwr writes sequentially) that you should take into consideration and depending on your specific application requirements they matter with a varying degree.

E.g. perhaps if your redo log files are small, or transactions redo amount is small, or archiving is turned off, or total archive store is small, you could use either Raid 1 or the striped+mirror variants (Raid 0+1 or Raid 10). Raid 5 requires at the very least three disks and an expensive controller. If you already have available a large raid 5 array based on say 10 disks or more, perhaps using that for redo log files together with other files could give satisfactory performance...

Regards,
/Fad Received on Fri Jan 05 2001 - 13:25:22 CST