raw device on IDE hard disk on RedHat Linux?

Last time I posted a message about raw device on Linux, the conclusion I got was that is's availabe on Redhat Linux 7 but only supports SCSI hard disk. Is there any way to make it work on IDE hard disk? "raw" is available on Redhat 6 as well, is there any major imporvement on Redhat 7 ? On RH 6 I used "raw" to bind a block device to /dev/raw/raw1 but the raw device is not quite usable. I used SCO UNIX on my PC in 1995 and raw device on IDE disk was well supported (once I create a division on disk there is a character device automatically created). Wasn't Linux a bit uncommercialised after so many years raw device is still a luxury to get? Carl Wu

Carl Wu wrote: > > Last time I posted a message about raw device on Linux, the conclusion I got > was that is's availabe on Redhat Linux 7 but only supports SCSI hard disk. > Is there any way to make it work on IDE hard disk? "raw" is available on > Redhat 6 as well, is there any major imporvement on Redhat 7 ? On RH 6 I > used "raw" to bind a block device to /dev/raw/raw1 but the raw device is not > quite usable. > I used SCO UNIX on my PC in 1995 and raw device on IDE disk was well > supported (once I create a division on disk there is a character device > automatically created). Wasn't Linux a bit uncommercialised after so many > years raw device is still a luxury to get? > > Carl Wu The raw utility is available from ftp://ftp.uk.linux.org:/pub/linux/sct/fs/raw-io/ and although I use it with SCSI, it is said to work with IDE as well. In fact, in the README, this is discussed I believe. There was another raw-device patch from SGI which was only for SCSI. That is however not how it is in linux 2.4. In linux 2.4 it is the former patch that i used, which works for IDE and SCSI (any block device).

"Carl Wu" <carlywu@yahoo.com> writes: > Last time I posted a message about raw device on Linux, the conclusion I got > was that is's availabe on Redhat Linux 7 but only supports SCSI hard disk. > Is there any way to make it work on IDE hard disk? "raw" is available on > Redhat 6 as well, is there any major imporvement on Redhat 7 ? RedHat released a 2.2.17 kernel for RHL7 that fixes a "raw" bug that could cause corruption. You should be using SCSI just for the ability to turn off the drive's on-board write re-ordering. To do otherwise could leave your database in an inconsistent state should a powerfail prevent the drive from writing all it's buffered data (you know, the data that the drive indicated was written when it really wasn't). > Wasn't Linux a bit uncommercialised after so many years raw device > is still a luxury to get? Linus was quite unconvinced of the necessity/utility of adding raw devices to Linux. -- Forte International, P.O. Box 1412, Ridgecrest, CA 93556-1412 Ronald Cole <ronald@forte-intl.com> Phone: (760) 499-9142 President, CEO Fax: (760) 499-9152 My GPG fingerprint: C3AF 4BE9 BEA6 F1C2 B084 4A88 8851 E6C8 69E3 B00B

Ronald Cole wrote in message ... > >> Wasn't Linux a bit uncommercialised after so many years raw device >> is still a luxury to get? > >Linus was quite unconvinced of the necessity/utility of adding raw >devices to Linux. > That's a common attitude of people who never use raw spaces. Sometimes UNIX buffering algorithms are not the best. The other main point about going raw is bypassing the UNIX buffers. Why save in an intermediate buffer when it's going to be put into shared memory? The shared memory usage of informix makes UNIX buffers irrelevant. But I guess you all know that... Perhaps some sort of NO_BUFFERING marker on file streams is an alternative. I understand (vaguely) that Informix NT can use _unbuffered_ NT files as efficiently as raw space, so it's recommended you use NT files for the convenience.

[This followup was posted to comp.databases.informix and a copy was sent to the cited author.] Excusing my ignorance, but we found that NT Informix was ugly slow on IO as opposed to a unix server. What did I miss brett > Perhaps some sort of NO_BUFFERING marker on file streams is an alternative. > I understand (vaguely) that Informix NT can use _unbuffered_ NT files as > efficiently as raw space, so it's recommended you use NT files for the > convenience. > > > >

bob wrote in message ... > >Excusing my ignorance, but we found that NT Informix was ugly slow on IO >as opposed to a unix server. What did I miss > Don't misunderstand me - you probably missed nothing! I'm not pushing the NT wheelbarrow. It seems to support 1/4 the number of users as an equivalent UNIX box for our applications. What I'm referring to is a few lines in the Informix manuals which say: The engine can be installed on NT with three choices of chunk storage: Files, Unbuffered NT files and raw spaces. It then goes on to say files are slower, but the unbuffered files are about the same speed as raw spaces so you should use unbuffered files because it's easier to manage. [invoke ancient hex rituals against NT now]

In the year of Our Lord Thu, 15 Feb 2001 09:31:40 +1100, "Andrew Hamm" <ahamm@sanderson.net.au> spake, saying: >bob wrote in message ... >> >>Excusing my ignorance, but we found that NT Informix was ugly slow on IO >>as opposed to a unix server. What did I miss >> >Don't misunderstand me - you probably missed nothing! I'm not pushing the NT >wheelbarrow. Uphill, eh? In the garden? :-)

Obnoxio The Clown wrote in message <3a8b0b55.122306717@130.133.1.4>... >In the year of Our Lord Thu, 15 Feb 2001 09:31:40 +1100, "Andrew Hamm" ><ahamm@sanderson.net.au> spake, saying: > >>Don't misunderstand me - you probably missed nothing! I'm not pushing the NT >>wheelbarrow. > >Uphill, eh? > >In the garden? > Too busy pushing the reset button