[Gluster-users] PageCache in Gluster Fuse and NFS clients

[Behrooz Shafiee]

Thanks for clarification Niels, but I am not yet clear if the NFS used to
mount GlusterFS is the same as stock NFS in the kernel tree (which uses
FS-Cache)?
My other question is that does NFS client of GlusterFS commits and blocks
for each write(not flush, just a write)? or does it use "weak cache
consistency model" allowing some buffering for writes.

Thanks,

On Tue, Apr 28, 2015 at 7:29 AM, Niels de Vos <ndevos at redhat.com> wrote:

> On Mon, Apr 27, 2015 at 02:56:14PM -0400, Behrooz Shafiee wrote:
> > Hello all,
> >
> >  I am a graduate student at University of Waterloo and working on
> > filesystem. I have two questions about GlusterFS that I will appreciate
> if
> > anyone can comment about.
> >
> > First, does GlusterFS FUSE client bypass Linux page cache mechanism(mount
> > with -o direct_io)? I have a machine with 64 GB of memory and when I
> write
> > a about 30GB of data into that machine (less than available memory) and
> > read it again, I see the reads are server through network, even though
> > those file could/are in the Linux page cache memory. It seems that's the
> > case both for NFS and FUSE clients.
>
> Direct IO is intended to not use caches where possible. What you
> describe above is the expected behaviour. You can find more details in
> 'man 2 open' for the O_DIRECT flag.
>
> The Linux kernel has many tunables for the page cache. You can find many
> of them under /proc/sys/vm/ , some are documented in 'man 5 proc'.
>
> > My second question is about FS-Cache and if GlusterFS does inode tree
> cache
> > on its FUSE clients or not. I know it exist in NFS but is the NFS client
> of
> > GlusterFS same as regular NFS?
>
> FS-Cache is a client-side option for some filesystems. The Linux kernel
> supports FS-Cache for its NFS-client. Some work has been done to add
> support for FS-Cache to FUSE, but that never became stable enough to get
> included in the upstream Linux kernel. Vimal Kumar did most of the work
> for that, and had a proof-of-concept. It will need updating for the
> current kernel version and some stability fixes. One of the email
> threads is here:
>
>     http://thread.gmane.org/gmane.comp.file-systems.gluster.devel/7877
>
> Vimal was planning to continue working on this, and your email is a
> great reminder to get a status update from him.
>
> Cheers,
> Niels
>



-- 
Behrooz
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gluster.org/pipermail/gluster-users/attachments/20150428/a2ecb9e4/attachment.html>