Messages in this thread | | | Date | Sat, 19 Dec 1998 10:13:01 +1300 | From | Chris Wedgwood <> | Subject | Re: mmap() is slower than read() on SCSI/IDE on 2.0 and 2.1 |
| |
On Fri, Dec 18, 1998 at 09:56:00PM +0100, Rik van Riel wrote:
> - the database stuff you mentioned
doesn't use mmap (well, I don't think it does) <insert raw device thread here>
> - a lot of tasks that access insanely large 'arrays', eg: > - (astronomical) image viewers
could be -- arguably, the applications might be able to be coaxed into reading forwards here
> - matrix solvers
can't comment
> - movie stuff when you watch an image sequence backwards > (sometimes done during editing, skips because of a faulty > read-ahead/behind algorithm are not tolerated)
my experience with this was with multi GB files (XFS on SGI vaults, a couple TB in total) and it bascially came down to raw IO speed, no kernel trickery here is going to help
smaller streams, maybe
> Why spend money on gobs of disks when you can put a simple > algorithm in the kernel that accomplishes the same effect?
I aggree with you -- but I'm not convinced having the kernel second guess the application is a good idea. <insert raw device thread here>
-cw
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |