Hi, On Mon, Feb 04, 2002 at 05:45:23PM +0100, Joe Radinger wrote: > > Stop! Before making comparisons, please look for the average > > _throughput_. It may happen that you have the _feeling_ that the machine > > is slower since _more_ data is transferred using ext3 and the CPU load > > is higher. > > would this mean that at the same HIGH systemload, ext3 is notable slower > than ext2, because disk-IO is expensive? No --- ext3 journal writes are sequential, so you can actually avoid seeking for those writes in some cases, and the eventual writeback can be deferred or indeed eliminated if the journal copy gets superceded. When ext3 is slower than ext2, it is usually because ext3 has stricter write ordering, so that you get some ordering constraints between processes which are not there in ext2. That only happens under extreme load: under most conditions all ext3 writes are completely asynchronous. ext3 also tends to push data to disk more quickly than ext2 if you use all the default settings. That's by design --- it means that after a crash, ext3 absolutely guarantees that you won't see stale data blocks show up in a file somewhere. You can loosen that guarantee by mounting the fs with "data=writeback" --- it's a compromise between data integrity and absolute performance. > i have some customers, that cannot switch to 2.4 yet, because they use > some custom network-modules, which are not available/compatible with > 2.4. they would like to use some journaling system, but they ant to > decide for themselfes [that means, that some of them would use ext3 and > other reiserfs] :( They are welcome to choose. Right now, though, all ext3 development is done on 2.4 only. Cheers, Stephen
