Re: openMPI vs openMosix
"Wes Wagner" <[email protected]> Mon, 25 Jun 2007 09:32:57 -0700
| Newsgroups | gmane.linux.cluster.openmosix.general |
|---|---|
| Message-ID | <[email protected]> |
--===============1981970303== Content-Type: multipart/alternative; boundary="----=_Part_25145_23499315.1182789177961" ------=_Part_25145_23499315.1182789177961 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Content-Disposition: inline Well, who knows, - the speed of light isn't actually a constant... so maybe we will have better luck 6 months from now. ;) -Wes On 6/25/07, Moshe Bar <[email protected]> wrote: > > absolutely. the issue of shm performance comes up every six months in > the openMosix mailing ilsts for these past six years, meanwhile the > underlying physics and simple queuing theory haven't changed much at > all. > > Moshe > > > > On 6/25/07, Wes Wagner <[email protected]> wrote: > > Is this not a classic, use the right tool for the right job issue? If > you > > need massive scalable symmetric multi-processing, using a large SMP > machine. > > If you can and need to do massive parallel multiple independent > processes > > that share nothing, use a cluster or batch processing system. > > > > If you can't afford a massive SMP system, redefine the problem.. ? > > > > -Wes > > > > > > On 6/25/07, Moshe Bar < [email protected]> wrote: > > > not only is the normal RAM access a full 3 to 4 orders of magnitude > > > faster, but also there is simulatenous access to cache lines > > > resolution, whereas a shared mem implementation has to relay on the > > > MMU, which only does 4096 bytes as a minimum, which vastly increases > > > the true and false sharing conflicts. > > > > > > To have a NIC help with that it would have to do something like RDMA, > > > with (at the least) MMU protection synced with the MMUs of all > > > connected nodes and remote interrupt capability. I doubt, however, > > > that such a dream NIC would help much in terms of performance. This > > > would be very close to the Altix solution, and witness their problems: > > > nodes have to be very close to each other (max 2 or 3 feet I think), > > > and > > > > > > On 6/25/07, Christian Leber < [email protected]> wrote: > > > > On Mon, Jun 25, 2007 at 10:41:24AM -0500, Moshe Bar wrote: > > > > > we used to have it. Shared memory without hardware support very > > > > > severely limits the types of apps that will actually scale. Even > sgi > > > > > in their altix machines (which are essentiallys shared memory > > > > > machines) have very limited scalability for a lot of apps, if not > > > > > most. > > > > > > > > What hardware features would be desireable to have in a NIC to be > able > > > > to implement shared memory halfway fast for Mosix on page level? > > > > > > > > Regards > > > > Christian Leber > > > > > > > > -- > > > > http://rettetdieti.vde-uni-mannheim.de/ > > > > > > > > > > > > > > > ------------------------------------------------------------------------- > > > > This SF.net email is sponsored by DB2 Express > > > > Download DB2 Express C - the FREE version of DB2 express and take > > > > control of your XML. No limits. Just data. Click to get it now. > > > > http://sourceforge.net/powerbar/db2/ > > > > _______________________________________________ > > > > openMosix-general mailing list > > > > openMosix-general-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org > > > > > > https://lists.sourceforge.net/lists/listinfo/openmosix-general > > > > > > > > > > > > > ------------------------------------------------------------------------- > > > This SF.net email is sponsored by DB2 Express > > > Download DB2 Express C - the FREE version of DB2 express and take > > > control of your XML. No limits. Just data. Click to get it now. > > > http://sourceforge.net/powerbar/db2/ > > > _______________________________________________ > > > openMosix-general mailing list > > > openMosix-general-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org > > > > > https://lists.sourceforge.net/lists/listinfo/openmosix-general > > > > > > > > > > > -- > > Wes Wagner > > > > Join a libertarian network of person-to-person lending on Prosper: > > > > > http://www.prosper.com/groups/group_home.aspx?group_short_name=freelibertarians&referrer=AiriusTorpora&utm_source=referrer-AiriusTorpora&utm_medium=referral-link&utm_content=join_my_group-160x33&utm_campaign=referrals-group > -- Wes Wagner Join a libertarian network of person-to-person lending on Prosper: http://www.prosper.com/groups/group_home.aspx?group_short_name=freelibertarians&referrer=AiriusTorpora&utm_source=referrer-AiriusTorpora&utm_medium=referral-link&utm_content=join_my_group-160x33&utm_campaign=referrals-group ------=_Part_25145_23499315.1182789177961 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Well, who knows, - the speed of light isn't actually a constant... so maybe we will have better luck 6 months from now. ;)<br><br>-Wes<br><br><br><div><span class="gmail_quote">On 6/25/07, <b class="gmail_sendername">Moshe Bar </b> <<a href="mailto:[email protected]">[email protected]</a>> wrote:</span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">absolutely. the issue of shm performance comes up every six months in <br>the openMosix mailing ilsts for these past six years, meanwhile the<br>underlying physics and simple queuing theory haven't changed much at<br>all.<br><br>Moshe<br><br><br><br>On 6/25/07, Wes Wagner <<a href="mailto:[email protected]"> [email protected]</a>> wrote:<br>> Is this not a classic, use the right tool for the right job issue? If you<br>> need massive scalable symmetric multi-processing, using a large SMP machine.<br>> If you can and need to do massive parallel multiple independent processes <br>> that share nothing, use a cluster or batch processing system.<br>><br>> If you can't afford a massive SMP system, redefine the problem.. ?<br>><br>> -Wes<br>><br>><br>> On 6/25/07, Moshe Bar < <a href="mailto:[email protected]">[email protected]</a>> wrote:<br>> > not only is the normal RAM access a full 3 to 4 orders of magnitude<br>> > faster, but also there is simulatenous access to cache lines<br> > > resolution, whereas a shared mem implementation has to relay on the<br>> > MMU, which only does 4096 bytes as a minimum, which vastly increases<br>> > the true and false sharing conflicts.<br>> > <br>> > To have a NIC help with that it would have to do something like RDMA,<br>> > with (at the least) MMU protection synced with the MMUs of all<br>> > connected nodes and remote interrupt capability. I doubt, however, <br>> > that such a dream NIC would help much in terms of performance. This<br>> > would be very close to the Altix solution, and witness their problems:<br>> > nodes have to be very close to each other (max 2 or 3 feet I think), <br>> > and<br>> ><br>> > On 6/25/07, Christian Leber < <a href="mailto:[email protected]">[email protected]</a>> wrote:<br>> > > On Mon, Jun 25, 2007 at 10:41:24AM -0500, Moshe Bar wrote: <br>> > > > we used to have it. Shared memory without hardware support very<br>> > > > severely limits the types of apps that will actually scale. Even sgi<br>> > > > in their altix machines (which are essentiallys shared memory <br>> > > > machines) have very limited scalability for a lot of apps, if not<br>> > > > most.<br>> > ><br>> > > What hardware features would be desireable to have in a NIC to be able <br>> > > to implement shared memory halfway fast for Mosix on page level?<br>> > ><br>> > > Regards<br>> > > Christian Leber<br>> > ><br>> > > --<br>> > > <a href="http://rettetdieti.vde-uni-mannheim.de/"> http://rettetdieti.vde-uni-mannheim.de/</a><br>> > ><br>> > ><br>> > ><br>> -------------------------------------------------------------------------<br>> > > This SF.net email is sponsored by DB2 Express <br>> > > Download DB2 Express C - the FREE version of DB2 express and take<br>> > > control of your XML. No limits. Just data. Click to get it now.<br>> > > <a href="http://sourceforge.net/powerbar/db2/"> http://sourceforge.net/powerbar/db2/</a><br>> > > _______________________________________________<br>> > > openMosix-general mailing list<br>> > > <a href="mailto:openMosix-general-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org"> openMosix-general-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org</a><br>> > ><br>> <a href="https://lists.sourceforge.net/lists/listinfo/openmosix-general">https://lists.sourceforge.net/lists/listinfo/openmosix-general</a><br>> > > <br>> ><br>> ><br>> -------------------------------------------------------------------------<br>> > This SF.net email is sponsored by DB2 Express<br>> > Download DB2 Express C - the FREE version of DB2 express and take <br>> > control of your XML. No limits. Just data. Click to get it now.<br>> > <a href="http://sourceforge.net/powerbar/db2/">http://sourceforge.net/powerbar/db2/</a><br>> > _______________________________________________ <br>> > openMosix-general mailing list<br>> > <a href="mailto:openMosix-general-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org">openMosix-general-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org</a><br>> ><br>> <a href="https://lists.sourceforge.net/lists/listinfo/openmosix-general"> https://lists.sourceforge.net/lists/listinfo/openmosix-general</a><br>> ><br>><br>><br>><br>> --<br>> Wes Wagner<br>><br>> Join a libertarian network of person-to-person lending on Prosper:<br>> <br>> <a href="http://www.prosper.com/groups/group_home.aspx?group_short_name=freelibertarians&referrer=AiriusTorpora&utm_source=referrer-AiriusTorpora&utm_medium=referral-link&utm_content=join_my_group-160x33&utm_campaign=referrals-group"> http://www.prosper.com/groups/group_home.aspx?group_short_name=freelibertarians&referrer=AiriusTorpora&utm_source=referrer-AiriusTorpora&utm_medium=referral-link&utm_content=join_my_group-160x33&utm_campaign=referrals-group </a><br></blockquote></div><br><br clear="all"><br>-- <br>Wes Wagner<br><br>Join a libertarian network of person-to-person lending on Prosper:<br> <br><a href="http://www.prosper.com/groups/group_home.aspx?group_short_name=freelibertarians&referrer=AiriusTorpora&utm_source=referrer-AiriusTorpora&utm_medium=referral-link&utm_content=join_my_group-160x33&utm_campaign=referrals-group"> http://www.prosper.com/groups/group_home.aspx?group_short_name=freelibertarians&referrer=AiriusTorpora&utm_source=referrer-AiriusTorpora&utm_medium=referral-link&utm_content=join_my_group-160x33&utm_campaign=referrals-group </a> ------=_Part_25145_23499315.1182789177961-- --===============1981970303== Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline ------------------------------------------------------------------------- This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/ --===============1981970303== Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline _______________________________________________ openMosix-general mailing list openMosix-general-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org https://lists.sourceforge.net/lists/listinfo/openmosix-general --===============1981970303==--