On Sun, Aug 17, 2003 at 02:21:48AM -0400, Tim McGrath wrote: > On Fri, 2003-08-15 at 21:47, Toad wrote: > > 3. DO NOT USE REISERFS3 (4's allocate on flush _might_ work better, but > > untested) for Freenet debug logs! The above was alleviated significantly > > by changing my log partition to ext2. So far messageSendTime's look > > fine, with ext2 and logLevel=debug, but it's only been up a few minutes. > > If they stay fine I can avoid spending $1000 on SCSI hardware :) > > > > Of course reiserfs is quite suitable for datastores, AFAIK. > > Just felt like pointing out something I read. Recently a bunch of > different FS's were benchmarked, and reiser3 was found to use the most > cpu out of all of them. Ext2 wasn't the best, but it WAS one of the > fastest out there, and used very little cpu. One thing you might try is
Reiserfsv3 is plenty fast for most things; it's just that its high CPU usage is a problem for the constantly-appending-small-fragments-at-speed usage pattern. I suspect its performance for Freenet datastores is quite respectable, especially compared to 2.4.X's ext3 without htree. > upgrading to ext3, which is basically painless if your kernel already > supports it - it uses less cpu than reiserfs, although it's *not* the > best on the list in anything IIRC. ext3 works a little differently than Right. The pattern with reiser3 was always more CPU, more performance in most things (especially metadata), given that CPU is not an issue. Usually it isn't. For Freenet debug level log files, it is. > ext2 in practice - most of the time, ext3 will wind up scheduling writes > to happen every five seconds or so. When you do large amounts of moving > data, ext3 doesn't perform very well however. Given that I prefer a > journalling filesystem to none at all, and ext3 performs 'good enough' > and is stable and reliable enough to be used, personally I prefer it. My > own node isn't the top of the pack, certaintly, but it performs > adequately. I do not need journaling for my log partition. > > So, basically reiser3 is unusuable by mostly anyone who doesn't want > high cpu usage for their filesystem. Hell, it nearly *crippled* my poor > demented 486 when I tried using it on it. But, that's a single case - > however it does demonstrate that the fs really does need a large amount > of cpu time just to function. Most of the time, it's fine, because CPU speed is a lot cheaper than hard disk speed. Except in applications which really stress its weak points. Such as high volume log files. > > If you need additional information, I'll go look up URLs and give you > any information you ask about my own stuff. > > Just my two cents, > Tim McGrath -- Matthew J Toseland - [EMAIL PROTECTED] Freenet Project Official Codemonkey - http://freenetproject.org/ ICTHUS - Nothing is impossible. Our Boss says so.
pgp00000.pgp
Description: PGP signature
