Re: [Moosefs-users] Performance about mfs mount

[Robert S. <rsa...@ne...>]

I have written quite a few emails on this list about this specific subject.

Fuse only seems to be a limitation if you use asynchronous I/O like 
sendfile(). Both Samba and Apache defaults to using sendfile().

At this stage my best guess is that this limit is due to the single 
socket between the mfsmount and mfsmaster instances that are re-used by 
everything.

In my production environment I have worked around the problem by having 
multiple mounts per server of the same volume and then load-balancing 
between the mounts.

The next limit seems to be that mfschunkserver will only process about 
10 requests at a time from all the different mfsmount instances. If you 
don't have a large number of mfschunkserver instances or you have 
mfschunkserver instances that control a large number of spindles then it 
can also become a bottleneck.

Robert

On 11/16/11 2:56 PM, Yezhou Fang wrote:
> I found out if I mount about twenty mount point in a server each has
> about 50 concurrent connections of read will get much better
> performance than only mount one which has about 500 connections . Does
> it because fuse is single process. It can not handle too many
> concurrent connections?
>
> ------------------------------------------------------------------------------
> All the data continuously generated in your IT infrastructure
> contains a definitive record of customers, application performance,
> security threats, fraudulent activity, and more. Splunk takes this
> data and makes sense of it. IT sense. And common sense.
> http://p.sf.net/sfu/splunk-novd2d
> _______________________________________________
> moosefs-users mailing list
> moo...@li...
> https://lists.sourceforge.net/lists/listinfo/moosefs-users