CNFS block size
Alex Kiernan
alexk at demon.net
Mon Sep 9 11:53:34 UTC 2002
Alex Kiernan <alexk at demon.net> writes:
> On Mon, 2002-09-09 at 01:55, Jeffrey M. Vinocur wrote:
> >
> > So I tried CNFS with raw devices on a redhat 2.4.18-3 system. It works.
> > But the RAID system would like all partitions to use the same block size.
> > (Every time a different block size is requested, it flushes the cache and
> > spams the logfile.) I see:
> >
> > #define CNFS_BLOCKSIZE 512 /* Unit block size we'll work with */
> >
> > What happens if I bump this up to 4k?
> >
> > (And should we make this a configure option?)
> >
>
> Its the allocation unit within the bitmap - the bigger you make it, then
> smaller your bitmap will be (for the same size of CNFS buffer) and the
> more space you will waste after every article.
>
> What it won't change (and I suspect this is what you think it might) is
> the size of the IO request issued - that will always be the size of an
> article, just aligned to a CNFS_BLOCKSIZE boundary.
>
Sorry missed a bit - the IO request is the article size for writes,
for reads you get an up front small read to determine the article
size, then a full blown read/mmap.
One useful thing to look into might be to encode the article size in
the token (assuming there's enough bits left).
--
Alex Kiernan, Principal Engineer, Development, THUS plc
More information about the inn-workers
mailing list