<div style="font-family: Helvetica; font-size: 13px; ">Hi Steffen,</div><div style="font-family: Helvetica; font-size: 13px; "><br></div><div style="font-family: Helvetica; font-size: 13px; "><blockquote type="cite" style="border-left-style: solid; border-left-color: rgb(0, 33, 98); color: rgb(0, 33, 98); border-top-width: 1px; border-right-width: 1px; border-bottom-width: 1px; border-left-width: 1px; margin-left: 0px; padding-left: 10px; padding-right: 0px; margin-right: 0px; "><span><div><div><blockquote type="cite" style="border-left-style: solid; border-left-color: rgb(0, 33, 98); color: rgb(0, 33, 98); border-top-width: 1px; border-right-width: 1px; border-bottom-width: 1px; border-left-width: 1px; margin-left: 0px; padding-left: 10px; padding-right: 0px; margin-right: 0px; "><div><div>I intend to build VM clusters for high-performance computing by</div><div>OpenNebula.</div></div></blockquote><div>We have a "virtual" HPC cluster, using the SGE,</div><div>and the nodes register themselves as SGE hosts during bootup.</div></div></div></span></blockquote><div><span><div><div><div><div><br></div><div>That sounds fine. I consider using SGE.</div><div>I have a little worry that using batch system is more complicated than</div><div>simple MPI environment for users. But it seems this is a right way.</div></div></div></div><div><br></div></span></div></div>
<div><div><blockquote type="cite" style="border-left-style: solid; border-left-color: rgb(0, 33, 98); color: rgb(0, 33, 98); border-top-width: 1px; border-right-width: 1px; border-bottom-width: 1px; border-left-width: 1px; margin-left: 0px; padding-left: 10px; padding-right: 0px; margin-right: 0px; "><span><div><div><blockquote type="cite" style="border-left-style: solid; border-left-color: rgb(0, 33, 98); color: rgb(0, 33, 98); border-top-width: 1px; border-right-width: 1px; border-bottom-width: 1px; border-left-width: 1px; margin-left: 0px; padding-left: 10px; padding-right: 0px; margin-right: 0px; "><div><div>But, in order to submit MPI job, it is necessary to know actual</div><div>compute-node (hostname, IP address, cores, etc...).</div><div>The number of VM of cluster can change on-demand by OpenNebula.</div><div>But maybe each VM cannot know change of member of cluster.</div></div></blockquote><div>Another possibility would be to run ganglia on the virtual nodes,</div><div>and query ganglia for the running hosts available for the MPI jobs.</div></div></div></span></blockquote><div> </div></div><div><span><div><div><div><div>Well, this is a good idea.</div><div><br></div><div>Thank you for your advice!</div></div></div></div></span></div><div><br></div><div>-- </div><div>Akihiko Ota</div></div>