Re: RE: Cook device vs Raw device

For raw disk, paging cache and the pagin mechanism are not used, the data is read or written directly between the device and the user address space. Therefore raw disk aviods the need to copy between kernel address space and user address space. This means that both reads and writes should seem the same performance improvement (ignoring the fancy write algorithms and pre-fetch algorithms etc). But as someone else pointed out if your buffer rates are high then the reads will not be going to disk, so you'll see the more benefits on writes. Paul Watson WF Software Ltd Tel: +44 1436 674729 Fax: +44 1436 678729 www.wfsoftware.com/informix # If you broke it, hide the evidence >>> "Reardon, Andrew J" <Andrew.Reardon@australia.boeing.com> 09/16/99 02:28AM >>> Correct if I'm wrong (I'm sure someone will :), but I always thought that raw disk only really starts giving you performance gains when it comes to doing lots of writes. Given this, the 1.5% gained below was due to the marginal performance gain from not having to go thru the block device interface for reads. Conversely, if William tried the test again, this time doing inserts/updates/deletes, he'll see very much improved figures. Andrew Reardon UNIX/Informix Administrator AeroSpace Support, Boeing Australia Ltd. Ph: +61 7 3306 3346 Mob: +61 0419 745 831 > ---------- > From: William Rice[SMTP:ricew@operamail.com] > Sent: Thursday, September 16, 1999 6:01 AM > To: William Rice > Cc: informix-list > Subject: RE: Cook device vs Raw device > > This is a thread that has always interested me. I have been reading this > group for awhile, and was concerned when I came to my current place of > employment > and discovered they were using filesystems to hold data. When we > got the oppurtunity to do some limited testing on raw vs cooked we > could only find at most a 1.5% difference between informix's performance > to raw disk vs. cooked. > We watched the following things: > iostat, > vmstat > wall time > > We were running AIX 4.3.2 on a 520 > Informix 7.31UC2 > > Only 2 disk drives, one of which was dedicated to informix data. > Under 2 Gigabytes of data in the whole database > > These processes were simple queries(no more than three tables joined) > and the loading of data into tables. > > The longest running process was somewhere around 15 minutes. > The processes tested caused the system to be heavily io bound for some > and CPU for others. > > From reading the newsgroup I got the impression that I could count on at > least > a 10% difference between cooked and raw, yet the tests I ran did not turn > out > that way. > > The data involved was literally identical. I was curious what factors > might > be causing the performance not to increase. > > Will > > ------------------------------------------------------------ > This e-mail has been sent to you courtesy of OperaMail, a > free web-based service from Opera Software, makers of > the award-winning Web Browser - http://www.operasoftware.com > ------------------------------------------------------------ > ********************************************************************** This email and any files transmitted with it are confidential and intended solely for the use of the individual or entity to whom they are addressed. If you have received this email in error please notify the system manager. This footnote also confirms that this email message has been swept by MIMEsweeper for the presence of computer viruses. www.mimesweeper.com **********************************************************************