<br>from: <a href="http://www.directron.com/expressguide.html">http://www.directron.com/expressguide.html</a><div><br></div><div>way more than you are using.</div><div><br></div><div>in a single direction PCI-E 4x has 2000MB/s<br>
<div><br></div><div><br><div><br></div><div><br><div class="gmail_quote">On Thu, Jun 16, 2011 at 11:02 AM, Matthew Zwier <span dir="ltr"><<a href="mailto:mczwier@gmail.com">mczwier@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
And by PCI I mean PCI Express.<br>
<div><div></div><div class="h5"><br>
On Thu, Jun 16, 2011 at 11:00 AM, Matthew Zwier <<a href="mailto:mczwier@gmail.com">mczwier@gmail.com</a>> wrote:<br>
> On Wed, Jun 15, 2011 at 10:17 PM, Drew from Zhrodague<br>
> <<a href="mailto:drewzhrodague@zhrodague.net">drewzhrodague@zhrodague.net</a>> wrote:<br>
>> Shouldn't you not share controllers with other arrays, and actually use<br>
>> multiple controllers per array? This is how we did things back in the<br>
>> day with old JBOD Suns -- stripe across disks and stripe across<br>
>> controllers. Modern architectures turn everything on their ears, so I<br>
>> could be missing something.<br>
><br>
> I doubt you're missing something, but we're out of slots in the<br>
> server, and out of space in the rack for a new one :) Did I mention<br>
> this is a scientific computing cluster for a small research group?<br>
> Funds and space are limited. Also, I can't correlate the performance<br>
> drop to a specific load on the other array.<br>
><br>
>> If it helps, I've found that a straight scp in Amazon's cloud is faster<br>
>> than an scp -C or compressed rsync. The results may be different with a<br>
>> hardware environment, but testing is the only way to be sure.<br>
><br>
> Yeah, tested this yesterday. Compression slows things down<br>
> considerably, Enumerating files (rsync) slows things down to kilobytes<br>
> per second. A typical user has a million or so tiny files and a few<br>
> hundred (10 GB - 200 GB) very large files. xfsdump over an<br>
> uncompressed ssh connection appears the way to go.<br>
><br>
> Interestingly, piping things through bar to measure throughput seems<br>
> to slow things down a *lot*, and in fact may be the greater reason for<br>
> the intermittent throughput -- I'm sustaining 4 MB/s - 50 MB/s,<br>
> seemingly correlated to what set of files xfsdump is working on, now<br>
> that I don't have bar in the pipe. I'm wondering if I need to do<br>
> something *really* dumb, like xfsdump -J - | xfsrestore -J -<br>
> /some/nfs/mount. That would seem to be going through one fewer<br>
> userland buffers.<br>
><br>
> Hmm...what's the bandwidth of a PCI bus?<br>
><br>
> MZ<br>
><br>
_______________________________________________<br>
wplug mailing list<br>
<a href="mailto:wplug@wplug.org">wplug@wplug.org</a><br>
<a href="http://www.wplug.org/mailman/listinfo/wplug" target="_blank">http://www.wplug.org/mailman/listinfo/wplug</a><br>
</div></div></blockquote></div><br></div></div></div>