Date: Sun, 8 Jul 2001 11:17:29 -0400 (EDT) From: "Albert D. Cahalan" <acahalan@cs.uml.edu> To: freebsd-stable@FreeBSD.ORG Cc: chopra@runbox.com Subject: Re: XFS (was: ReiserFS (was: JFS (was: The FreeBSD core team needs your help)))a Message-ID: <200107081517.f68FHT2209183@saturn.cs.uml.edu>
next in thread | raw e-mail | index | archive | help
Munish Chopra writes: > On Sun, Jul 08, 2001 at 02:36:01PM +0930, Greg Lehey wrote: >> On Sunday, 8 July 2001 at 2:22:23 +0200, Munish Chopra wrote: >>> I'm still more impressed by XFS (on paper), but I've never run it on >>> Linux myself. From those that have though, I hear it's very good, and >>> the SGI folks seem to be putting a lot of work into key areas. Some of >>> the kiddie diseases are wearing off apparently, and it's actually >>> maturing nicely. >> >> Agreed, it seems like quite a good design to me too. I have heard >> recently, however, that they needed to make a number of changes to the >> Linux kernel in order to get it to perform acceptably, and that Linux >> may not accept these changes. I don't know any details, however. > > Well, let's see. IIRC, this was because of the Linux VFS layer, which > has been ripped out, re-done, re-implemented, and just generally changed > many many times. Often the changes weren't minor, and coupled with some > of the other stuff going on in the kernel, I'm sure it would break > things (and it did breat many things). The VFS layer hasn't been ripped out since 2.1.44, long ago. Recent changes are mostly related to SMP. XFS places heavy demands on the VM/IO systems. For example, it does not choose the on-disk location for a file until the file must be placed on disk. This is good for performance, but can cause trouble when memory is low. To write out data, XFS may need to have memory. When the data is being written out to free up memory, this hurts. > Then, there've been some more or less behind-the-scenes fights about > what kind of changes will be accepted for Linux. I remember a few > myself, though these related to IBM. Considering SGI and how much it's > betting on Linux, this type of stuff is becoming something of a brick > wall for them. No kidding. Linus isn't going to take the first crude hack that comes along. Long-term maintenance, portability, and performance have to be considered when changing core kernel interfaces. Do you think the FreeBSD commiters would accept the first crude hack without any argument? To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200107081517.f68FHT2209183>