On 5/19/2010 1:24 PM, Bill Bogstad wrote:
> http://unfs3.sourceforge.net/
>   

Thanks.  I had checked out the Linux userspace NFS daemon but missed
this one.

> in Google.  It claims to run under Solaris and has had updates in the
> last 1-2 years.   No idea about reliability, performance, or
> re-exporting.   Since all data from your new 7210 will flow through
> your old 7210, performance is definitely going to be worse for those
> filesystems.   If you take this approach (and I'm not recommending
> it), you will probably want to dedicate
>  at least some of the Ethernet ports on both machines for dedicated
> machine-machine connections just for the NFS traffic.
> Using Ethernet jumbo frames for those links might be a good idea as well.
>   

I'm trying it out now.  It does allow re-exporting.  My initial sanity
test using a VM as the relay (time cat 821MB-file > /dev/null) shows
that the re-exported system is about 70% of direct mount speed, with the
iostat measured NFS reads/s peaking at 60%.  This clearly warrants
further testing. 

>From my experience with VMWare, it's latency that will kill me -- VMware
is all about the IOps and latency and uses relatively little bandwidth,
so I'm going to concentrate on reliability and latency.

If anyone has an idea of what sharp edges I might encounter trying this
out, please let me know so I can add that to my validation tests of the
idea.

Many NICs is a good idea, also.  I have 8 on each 7210, so I can
dedicate some to this link.

Thanks,

--
Dewey

_______________________________________________
bblisa mailing list
[email protected]
http://www.bblisa.org/mailman/listinfo/bblisa

Reply via email to