On Wed, 2002-03-20 at 20:42, civileme wrote: > actually hdparm -t /dev/hdx is more informative. The capital T > basically measures the bandwidth of your memory. > > And if you are using software RAID, then it is less than informative. > It works OK for RAID1 and 4 but not so well for 0 or 5, especially not > for 0, where it is possible to exceed the speed of a single disk. > > But my measurements on creation and updating show XFS on a par or > slightly better than ext2 which is itself 50% faster than ext3. (Yes, > ext3 is like 66% the speed of ext2).
Perhaps this is true under your conditionals, but under what ext3 journalling mode were your tests run? There are three; Nonjournalling, metadata mode, and data mode. If you are going to bench ext2 against ext3, the tests should include results from ext3 in nonjournalling mode. If you are going to bench ext3 against any other journaling fs, the tests should include results from ext3 in metadata journalling mode. (Since all the other journaling fs's only support metadata journalling.) If you want to prove how slow ext3 can truly be, you should bench it in it's maximum data protection option, which is data journalling mode. As far as I know, ext3 is the only FS that supports this level of operational reliability. This mode is of course for those that do not consider speed as a sole requisite for filesystem analysis. There are such people out there. I don't mind benchmarks, in fact I burn the wires between myself and Dr Tom, but I run ext3 and I haven't seen the kind of speed differences indicated in your tests. I can't begin to make any comparisons beyond that (with XFS or otherwise) because I havent seen the specifics concerning ext3 journalling mode under your tests. I also wonder how small an XFS partition could be. ;) > Civileme Late R On LX _________________________________________________________ Do You Yahoo!? Get your free @yahoo.com address at http://mail.yahoo.com
Want to buy your Pack or Services from MandrakeSoft? Go to http://www.mandrakestore.com