Re: [fuse-devel] FUSE write performance

[Kevin F. <Kev...@pn...>]

On Thu, 2010-10-07 at 02:00 -0700, Miklos Szeredi wrote:
> On Tue, 5 Oct 2010, neutron wrote:
> > Hi all,
> > 
> > Got a question with regard to fuse performance.
> > 
> > I developed a fuse-based filesystem (fuse 2.7.4).  But now the write
> > throughput at this filesystem seems pretty low.
> > 
> > I ran a test with 8 processes on a 8-core machine (redhat 2.6.18) with
> > 6GB memory.  Each process opens and writes 1GB data in 1MB chunks to a
> > separate file in my fuse filesystem.  To measure the overhead of fuse
> > internally, my filesystem ignores the writing  in fuse_write(), and
> > returns directly without performing any writes.   In this test, the
> > aggregated write bandwidth of the 8 processes is only about 240 MB/s.
> > 
> > Any comments on how to improve write performance? Thanks!
> 
> Install a newer kernel (>=2.6.26), a newer version of libfuse
> (>=2.8.0) and use the 'big_writes' option.

What is the possibility to run a 2.6.26 fuse kernel module on an older
kernel? We still have a lot of rhel5 boxes and have been looking at a
way to do large writes without direct_io, which we are using now.

Thanks,
Kevin

> 
> Alternatively if you have to stick with the older kernel you can use
> '-odirect_io' to get bigger write requests, but that solution has
> other side effects.
> 
> Thanks,
> Miklos
> 
> ------------------------------------------------------------------------------
> Beautiful is writing same markup. Internet Explorer 9 supports
> standards for HTML5, CSS3, SVG 1.1,  ECMAScript5, and DOM L2 & L3.
> Spend less time writing and  rewriting code and more time creating great
> experiences on the web. Be a part of the beta today.
> http://p.sf.net/sfu/beautyoftheweb
> _______________________________________________
> fuse-devel mailing list
> fus...@li...
> https://lists.sourceforge.net/lists/listinfo/fuse-devel