Roy Sigurd Karlsbakk schrieb:
> > Right now the filesystem block size on linux is restricted to the
> > machine page size. Fixing this has been on the todo list for a
> > long time now, but other projects keep getting given higher priority.
> > The first thing to appear would be support for smaller block sizes,
> > not bigger ones - the problem in linux is that the kernel was really
> > not designed to provide chunks of memory larger than 1 page with any
> > reliability (i.e. there is a good chance it will fail).
> > Anyway, having said that, xfs will do larger I/O than this, files should
> > get layed out contiguous on disk, and both the read and write path will
> > end up issuing larger requests to the scsi devices.
> ok. So the fs block size really doesn't matter - is that it? I'm trying to
> set up streaming of multi-gigabyte files from a file system, and it's all
> too slow. Even with a RAID-0 with 5 drives (SCSI-3/160MBps/10k spin), I
> cannot get the speed higher than around 30 on ext2.
Do you say 30M/s. That's not very good. If it is raw read speed, there
is something wrong. I get the following on a DELL PowerEdge 1400, 4x
U160 SCSI drives and software RAID5 on it:
[root@install /home]# ls -la TEST.bin
-rw-r--r-- 1 root root 4096000000 Nov 13 16:11 TEST.bin
[root@install /home]# time dd if=TEST.bin of=/dev/null bs=16k
250000+0 records in
250000+0 records out
The md is:
md4 : active raid5 sdd7 sdc7 sdb7 sda7
21442176 blocks level 5, 64k chunk, algorithm 0 [4/4] [UUUU]
So this is 108M/s raw read speed.
> Do you or anyone else have some idea of when the TUX/XFS potch
> incompatibility problem will be solved? I need Tux (the in-kernel web
> server from RedHat) in this project as well...
> Roy Sigurd Karlsbakk, MCSE, MCNE, CLS, LCA
> Computers are like air conditioners.
> They stop working when you open Windows.