Re: How To create raw devices in Red Hat 5.2
First confessing that I know nothing in particular about Red Hat, you should be able to do the same thing for Informix. Digging back into my distant past, I seem to recall that you do everything you would do for a normal file system except to mount it. The net effect is that I/O to that disk will bypass the UNIX buffer cache, as mentioned. The main advantage of raw vs. cooked is one of security of the data. If data is buffered by UNIX, there is no way for Informix to know what has been physically placed on the disk vs. still in the buffer cache, and in the event of a system crash the restorability of the database becomes problematic. On the other hand, it is definitely possible that having disk go to cooked (and therefore buffered) devices could improve performance somewhat. But that scenario isn't all that common given sufficient other resources (i.e., shared memory) to allow Informix to do its thing to raw disk. HTH, David manmtn@mat.net (ManMtn) on 02/15/99 05:40:34 PM Please respond to manmtn@mat.net (ManMtn) To: informix-list@iiug.org cc: (bcc: David Coburn/ISG/WVUS/WorldVision) Subject: Re: How To create raw devices in Red Hat 5.2 In article <7a9b9f$gtg$1@nnrp1.dejanews.com>, neil_rutherford@yahoo.com wrote: I have not done much RDBMS in a while, but I know with Sybase, you can spacify a partition that has not been allocated by a filesystem. I suspect all you would have to do is have the partition created on the hard drive, and then point the database at it. The raw linux disk drivers should work, as long as Informix knows how to deal with them. Brad > I asked the same question a month or so ago and put it mildly, > i believe you can create raw devices.... known as special character > devices.... still don't know how, yet, i've read in the release > notes that the kernel doesn't raw devices buffering or none-buffering ( i > can't remember ), so they said they don't support it, due to the fact that > there will be no performance gains over fried (cooked) files. > Does that make any sense? > Therefore cooked files has the same performance as raw disk. > > In article <36C76008.2DF2319B@weideneder.de>, > Stefan Weideneder <stefan@weideneder.de> wrote: > > Ricardo wrote: > > > > > > Hi, > > > > > > Does any one knows how to create a raw device for a Informix chunk? > > > > > > TIA > > > NB > > > > Hi Ricardo, > > > > are you sure that Informix supports "raw devices" for your > > Linux workstation ? Far as I know there's no way to create > > raw devices - sure, you can try by using "mknod" but, I guess > > your system will tell you the same as my "SuSe" Linux. > > > > Anyhow, the I/O performs pretty good. I got a throughput > > of about 4MB/sec for random write operations per disk. > > That's much more than on other platforms. > > > > Best regards > > > > Stefan Weideneder > > > > -----------== Posted via Deja News, The Discussion Network ==---------- > http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
LINUX doesn't support the method you described below, it's clearly stated that there isn't any performance gains with raw over cooked in the kernel up to 2.0.36, i'm not sure if this is true as of 2.2.x. The release notes with INFORMIX (Linux) mentions this! In article <7aat4r$dj9$1@news.xmission.com>, "David Coburn" <dcoburn@worldvision.org> wrote: > > First confessing that I know nothing in particular about Red Hat, you > should be able to do the same thing for Informix. Digging back into my > distant past, I seem to recall that you do everything you would do for a > normal file system except to mount it. The net effect is that I/O to that > disk will bypass the UNIX buffer cache, as mentioned. The main advantage > of raw vs. cooked is one of security of the data. If data is buffered by > UNIX, there is no way for Informix to know what has been physically placed > on the disk vs. still in the buffer cache, and in the event of a system > crash the restorability of the database becomes problematic. On the other > hand, it is definitely possible that having disk go to cooked (and > therefore buffered) devices could improve performance somewhat. But that > scenario isn't all that common given sufficient other resources (i.e., > shared memory) to allow Informix to do its thing to raw disk. > > HTH, > > David > > manmtn@mat.net (ManMtn) on 02/15/99 05:40:34 PM > > Please respond to manmtn@mat.net (ManMtn) > > To: informix-list@iiug.org > cc: (bcc: David Coburn/ISG/WVUS/WorldVision) > Subject: Re: How To create raw devices in Red Hat 5.2 > > In article <7a9b9f$gtg$1@nnrp1.dejanews.com>, neil_rutherford@yahoo.com > wrote: > I have not done much RDBMS in a while, but I know with Sybase, you can > spacify a > partition that has not been allocated by a filesystem. I suspect all you > would have to do is have the partition created on the hard drive, and then > point the database at it. The raw linux disk drivers should work, as long > as Informix knows how to deal with them. > > Brad > > I asked the same question a month or so ago and put it mildly, > > i believe you can create raw devices.... known as special character > > devices.... still don't know how, yet, i've read in the release > > notes that the kernel doesn't raw devices buffering or none-buffering ( i > > can't remember ), so they said they don't support it, due to the fact > that > > there will be no performance gains over fried (cooked) files. > > Does that make any sense? > > Therefore cooked files has the same performance as raw disk. > > > > In article <36C76008.2DF2319B@weideneder.de>, > > Stefan Weideneder <stefan@weideneder.de> wrote: > > > Ricardo wrote: > > > > > > > > Hi, > > > > > > > > Does any one knows how to create a raw device for a Informix chunk? > > > > > > > > TIA > > > > NB > > > > > > Hi Ricardo, > > > > > > are you sure that Informix supports "raw devices" for your > > > Linux workstation ? Far as I know there's no way to create > > > raw devices - sure, you can try by using "mknod" but, I guess > > > your system will tell you the same as my "SuSe" Linux. > > > > > > Anyhow, the I/O performs pretty good. I got a throughput > > > of about 4MB/sec for random write operations per disk. > > > That's much more than on other platforms. > > > > > > Best regards > > > > > > Stefan Weideneder > > > > > > > -----------== Posted via Deja News, The Discussion Network ==---------- > > http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own > > -----------== Posted via Deja News, The Discussion Network ==---------- http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own