> > I have two 500 GB SATA II drives. My programs do lots of file I/O and can > > generate files as large as 15-20 GB in some of my scientific applications. > > You could also consider xfs, it's very god for large files. Even better, > test your application with both types of filesystem, and then decide.
Testing is definitely the best way to reach a decision. About a year ago, I tested ext3, xfs, and reiserfs for our systems. We do molecular dynamics simulations on clusters, and that means that we have several clients appending to large files (>10GB) simultaneously. I've discovered that xfs came out fastest (and with a relatively low load on the NFS server), ext3 was second, and reiserfs clearly the weakest. However, when I repeated a few of the tests on a local file system, that picture changed drastically: reiserfs was more or less on par with xfs, with ext3 being way behind. So, it definitely depends on the details of your application. As for the stability, the data I have is much "softer" (read: anecdotal). We've had a few crashes with reiserfs, including some data loss. At one point, a rebuild_tree failed and left a disc with no file names and only a limited directory hierarchy. While xfs seems to be more stable, it shows some problems as well (no data loss so far). In my experience, ext3 has been rock solid. However, since fs crashes and data loss are such rare events (at least, they should be), it is quite hard to gather reliable data, and the signal-to-noise ratio tends to be low. Besides, reiserfs takes a long time to mount; and mkfs.xfs is incredibly fast. Of course, creating files systems (and even mounting them) is a rare occurrence, so these points should not influence your decision unless everything else is equal. A. -- Ansgar Esztermann Researcher & Sysadmin http://www.mpibpc.mpg.de/groups/grubmueller/start/people/aeszter/index.shtml
pgp80ntxKYIPN.pgp
Description: PGP signature