[OpenAFS] Tips for increasing throughput

Wed, 31 Aug 2011 13:43:04 -0700

--0016e64aff62f20d2404abd32d38
Content-Type: text/plain; charset=ISO-8859-1

We're using 1.4.14 on the file servers (Ubuntu Natty) and 1.4.12 on the
clients (Ubuntu Lucid). We don't know many details about the network
topology as our servers are sitting on EC2, though as I stated previously,
our scp benchmarks are significantly faster.

The fileserver configuration is the default from the repository:
/usr/lib/openafs/fileserver -p 23 -busyat 600 -rxpck 400 -s 1200 -l 1200 -cb
65535 -b 240 -vc 1200

On Wed, Aug 31, 2011 at 6:01 AM, Matt W. Benjamin <matt@linuxbox.com> wrote:

> Hi,
>
> I think some more detail on the network/network path would be helpful.
>
> Regards,
>
> Matt
>
> ----- "Ken Elkabany" <Ken@Elkabany.com> wrote:
>
> > Hi,
> >
> >
> > On a network capable of scp-ing files between machines at 60MB/sec, we
> > are only able to achieve 2-3MB/sec of throughput when using AFS. We've
> > been conducting tests on 300MB files, by copying them from the /afs/*
> > mount to the local filesystem with memcache enabled. In production, we
> > will not be using memcache, but we wanted to eliminate the overhead of
> > the on-disk cache manager. We've tried tweaking our options as
> > follows: "-verbose -nosettime -memcache -chunksize 20 -stat 2800
> > -daemons 5 -volumes 128". The most important option is the chunksize,
> > which did not have a measurable effect. Any other tips?
> >
> >
> > Thanks!
> > Ken
>
> --
>
> Matt Benjamin
>
> The Linux Box
> 206 South Fifth Ave. Suite 150
> Ann Arbor, MI  48104
>
> http://linuxbox.com
>
> tel. 734-761-4689
> fax. 734-769-8938
> cel. 734-216-5309
>

--0016e64aff62f20d2404abd32d38
Content-Type: text/html; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable

<br></div><div>The fileserver configuration is the default from the reposit=
ory: /usr/lib/openafs/fileserver -p 23 -busyat 600 -rxpck 400 -s 1200 -l 12=
00 -cb 65535 -b 240 -vc 1200</div><div><br></div><div><div class=3D"gmail_q=
uote">

On Wed, Aug 31, 2011 at 6:01 AM, Matt W. Benjamin <span dir=3D"ltr">&lt;<a =
href=3D"mailto:matt@linuxbox.com">matt@linuxbox.com</a>&gt;</span> wrote:<b=
r><blockquote class=3D"gmail_quote" style=3D"margin:0 0 0 .8ex;border-left:=
1px #ccc solid;padding-left:1ex;">

Hi,<br>
<br>
I think some more detail on the network/network path would be helpful.<br>
<br>
Regards,<br>
<br>
Matt<br>
<div><div></div><div class=3D"h5"><br>
----- &quot;Ken Elkabany&quot; &lt;Ken@Elkabany.com&gt; wrote:<br>
<br>
&gt; Hi,<br>
&gt;<br>
&gt;<br>
&gt; On a network capable of scp-ing files between machines at 60MB/sec, we=
<br>
&gt; are only able to achieve 2-3MB/sec of throughput when using AFS. We&#3=
9;ve<br>
&gt; been conducting tests on 300MB files, by copying them from the /afs/*<=
br>
&gt; mount to the local filesystem with memcache enabled. In production, we=
<br>
&gt; will not be using memcache, but we wanted to eliminate the overhead of=
<br>
&gt; the on-disk cache manager. We&#39;ve tried tweaking our options as<br>
&gt; follows: &quot;-verbose -nosettime -memcache -chunksize 20 -stat 2800<=
br>
&gt; -daemons 5 -volumes 128&quot;. The most important option is the chunks=
ize,<br>
&gt; which did not have a measurable effect. Any other tips?<br>
&gt;<br>
&gt;<br>
&gt; Thanks!<br>
&gt; Ken<br>
<br>
</div></div>--<br>
<br>
Matt Benjamin<br>
<br>
The Linux Box<br>
206 South Fifth Ave. Suite 150<br>
Ann Arbor, MI =A048104<br>
<br>
<a href=3D"http://linuxbox.com" target=3D"_blank">http://linuxbox.com</a><b=
r>
<br>
tel. <a href=3D"tel:734-761-4689" value=3D"+17347614689">734-761-4689</a><b=
r>
fax. <a href=3D"tel:734-769-8938" value=3D"+17347698938">734-769-8938</a><b=
r>
cel. <a href=3D"tel:734-216-5309" value=3D"+17342165309">734-216-5309</a><b=
r>
</blockquote></div><br></div>

--0016e64aff62f20d2404abd32d38--