<div>Yes, in fact Score libraries had nice results in MPI over Ethernet in a couple of clusters I've seen... in fact, by the way implemented by your company, in collaboration with a HW vendor. As far as I remember an efficiency of around 67-70% in HPL is manageable with Score MPI, if I remember well.
</div>
<div> </div>
<div>Problem was, as far as I remember, only specific versions of x86/ x86_64 linux kernel were supported, don't know if this is gonna change in the next future.</div>
<div> </div>
<div> </div>
<div> </div>
<div><span class="gmail_quote">2007/7/13, John Hearns <<a href="mailto:john.hearns@streamline-computing.com">john.hearns@streamline-computing.com</a>>:</span>
<blockquote class="gmail_quote" style="PADDING-LEFT: 1ex; MARGIN: 0px 0px 0px 0.8ex; BORDER-LEFT: #ccc 1px solid">Mark Hahn wrote:<br>>> Anyway, hop latency in Ethernet is most of times just peanuts in terms of<br>
>> latency compared to TCP/IP stack overhead...<br>><br>> unfortunately - I'm still puzzled why we haven't seen any open,<br>> widely-used,<br>> LAN-tuned non-TCP implementation that reduces the latency. it should be
<br>> possible to do ~10 us vs ~40 for a typical MPI-over-Gb-TCP.<br><br>Well, the SCore impementation which we install on all our clusters does<br>just this.<br><a href="http://www.pccluster.org">www.pccluster.org</a>
<br><br>In fact, we have one 500 machine cluster which (at the time of install)<br>ranked 167 in the Top 500 and achieved a very high efficiency.<br>All connected with gigabit ethernet only.<br><a href="http://www.streamline-computing.com/index.php?wcId=76&xwcId=72">
http://www.streamline-computing.com/index.php?wcId=76&xwcId=72</a><br><br><br><br><br>--<br> John Hearns<br> Senior HPC Engineer<br> Streamline Computing,<br> The Innovation Centre, Warwick Technology Park,
<br> Gallows Hill, Warwick CV34 6UW<br> Office: 01926 623130 Mobile: 07841 231235<br></blockquote></div><br>
!DSPAM:4697aef672732020149523!