Christoph Hellwig wrote:
> On Wed, Jan 24, 2001 at 01:20:01AM +0000, David Wragg wrote:
> > One possibility is that ext2 is the simply the wrong filesystem for my
> > application, and any other applications that needs guaranteed
> > sustained write bandwidth.  The obvious alternative is XFS.  However,
> > early releases of Linux XFS will not feature the guaranteed rate I/O
> > features, so that's not likely to be an option for quite a while yet.
> 
> I think the XFS feature you really want for this kind of workload 
> (besides GRIO) is the delayed allocation, which means writing to the
> pagecache without allocating a block on disk.
> XFS/Linux already has this feature, and I hope we'll get it as generic
> feature in 2.5.
> 
>       Christoph

What XFS also has is an ioctl to preallocate disk space, there is very
little documentation on this (none), but if you look in the file
cmd/xfstests/src/randholes.c you will see an ioctl like this:

struct flock64 fl;
fl.l_start = offset;
fl.l_len = blocksize;
fl.l_whence = 0;

if (ioctl(fd, XFS_IOC_RESVSP64, &fl) < 0) {

This is preallocating disk blocks for file data. Of course, it only
works if you know file size in advance. We should also have direct
I/O in the next month or so.

We also have the realtime subvolume - which is a little tricky to setup,
but basically you can set aside a chunk of disk space as realtime (a
different partition), the only data which goes into this space is file
data for files marked realtime (another ioctl call). Files in the realtime
subvolume are allocated differently - they are allocated in fixed size
chunks - usually large, using an allocator which gives out space using
a binary chop of the disk - which while wasteful, means file fragmentation
is very hard to achieve.

This will really come in handy with direct I/O, and is not currently
really being tested (i.e. I don't know if it is useable right now - let
me know if you are interested). I think we need more documentation on
some of these features of XFS.

GRIO is a different kettle of fish entirely. Most of the work in GRIO
is not in xfs, it is in the disk drivers being capable of doing
bandwidth allocation, and being able to follow a request all the way
down from a file descriptor to the block layer to keep track of the
bandwidth allocation it is part of. A drive (or drive subsystem) is specified
as having an available bandwidth. Applications can reserve bandwidth
(X Mbytes/second for Y seconds), and they either get the bandwidth
or they do not. Requests which are part of a bandwidth allocation are
given priority over other requests by the driver layers.

Steve



-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [EMAIL PROTECTED]

Reply via email to