Messages in this thread | | | Date | Tue, 06 Jan 1998 10:00:38 -0600 | From | "Dan A. Dickey" <> | Subject | Re: lack of raw disk devices |
| |
Peter Monta wrote:
> > Okay. I've seen this dicussion many times before. I need to know now. What the > > hell is the difference a RAW block device and a (err.... NONRAW?) regular block > > device? > > A raw device doesn't go through the buffer cache.
Another critical consideration when using raw devices is that if you issue a read() for 32K, the read is done in one 32K chunk.Same for write()s. I'm not sure how Linux handles this, but it would seem fairly critical for backing up to tapes...so much so, that I'm sure Linux must handle it some how. Can someone please inform me? Also, how does it handle read/writes of a non-block size? When I wrote tape drivers, lots of people would like to write wierd size records - say, 444 bytes or some such. ???
When I need to move filesystems around on other unix machines, I typically dd the raw device, about 64MB per read/write. This goes much faster than going through the block device. I haven't had to do this on Linux yet, as I only use Linux for test machines, not in a production environment. So, how does Linux handle large read/write requests? I can't believe that it doesn't as of 2.1.77 - I'm almost so surprised to find out that it doesn't have raw devices that I might find the time to go look at sources. -Dan
-- Dan A. Dickey http://www.transition.com/ mailto:ddickey@transition.com
begin: vcard fn: Dan Dickey n: Dickey;Dan org: Transition Networks, Inc. adr: 6475 City West Parkway;;;Eden Prairie;MN;55344;USA email;internet: ddickey@transition.com title: Sr. Software Engineer tel;work: 612-941-7600 tel;fax: 612-941-2322 x-mozilla-cpt: ;0 x-mozilla-html: TRUE version: 2.1 end: vcard
| |