[mvapich-discuss] Announcing the release of MVAPICH2 1.0-beta

Dhabaleswar Panda panda at cse.ohio-state.edu
Tue Jul 31 10:46:02 EDT 2007


Hi Eric, 

> Thank you again for you and your group's hard work on this software.

Thanks for your compliment.

> I'll start by saying that I know I should move over to OpenFabrics and Gen2,
> but as we've discussed previously, this isn't currently a viable option for
> reasons that are outside the scope of this forum. With that said...

No problem. 

In fact, for this Beta release, we focused primarily on OpenFabrics
Gen2 and uDAPL because most of the users now are using these
interfaces. We have not done any testing on VAPI yet and have plans to
do that this week. Thanks for identifying these compilation snags. We
added some new functionalities and it appears that some of the things
are broken for VAPI interfaces. We will take care of these and make
appropriate fixes.

We will get back to you with these fixes soon. 

Thanks, 

DK

> A few compilation snags with MVAPICH2-1.0-beta on the VAPI flavor:
> 
> (1) In src/mpid/osu_ch3/channels/mrail/src/rdma/ch3_read_progress.c line 146
> :
> 
> type = MPIDI_CH3I_MRAILI_Cq_poll(v_ptr, NULL, 0, is_blocking)
> 
> calls with four arguments; the VAPI version ( defined in
> src/mpid/osu_ch3/channels/mrail/src/vapi/mpidi_ch3_rdma_post.h ) has only
> the first three arguments. I imagine this is just a missing #ifdef switch
> ...
> 
> (2) in src/mpid/osu_ch3/channels/mrail/src/vapi/rdma_iba_1sc.c lines 151-156
> :
> 
> if (SMP_INIT)
> {
>     /*correspoding post has not been issued */
>     flag = 0;
>     break;
> }
> 
> These lines appear to have migrated here from somewhere else in the code
> (perhaps  the function immediately above it.) The variable flag is undefined
> at this point, and there's a break statement without a loop to break out
> of...
> 
> By no means a tested fix, but removing the last argument from the issue
> mentioned in (1) and commenting out the offending lines in (2) appears to
> allow the VAPI channel to compile and run (benchmarks, in-house tools)
> successfully. I haven't been able to get logging working, but that is
> another discussion.
> 
> Your thoughts?
> 
> Thanks again!
>  Eric Borisch
> 
> On 7/26/07, Dhabaleswar Panda <panda at cse.ohio-state.edu> wrote:
> > The MVAPICH team is pleased to announce the availability of
> > MVAPICH2-1.0-beta with the following NEW features:
> >
> > - Message coalescing support to enable reduction of per Queue-pair
> >   send queues for reduction in memory requirement on large scale
> >   clusters. This design also increases the small message messaging
> >   rate significantly. Available for Open Fabrics Gen2-IB.
> >
> > - Hot-Spot Avoidance Mechanism (HSAM) for alleviating
> >   network congestion in large scale clusters. Available for
> >   Open Fabrics Gen2-IB.
> >
> > - RDMA CM based on-demand connection management for large scale
> >   clusters. Available for OpenFabrics Gen2-IB and Gen2-iWARP.
> >
> > - uDAPL on-demand connection management for large scale clusters.
> >   Available for uDAPL interface (including Solaris IB implementation).
> >
> > - RDMA Read support for increased overlap of computation and
> >   communication. Available for OpenFabrics Gen2-IB and Gen2-iWARP.
> >
> > - Application-initiated system-level (synchronous) checkpointing in
> >   addition to the user-transparent checkpointing. User application can
> >   now request a whole program checkpoint synchronously with BLCR by
> >   calling special functions within the application. Available for
> >   OpenFabrics Gen2-IB.
> >
> > - Network-Level fault tolerance with Automatic Path Migration (APM)
> >   for tolerating intermittent network failures over InfiniBand.
> >   Available for OpenFabrics Gen2-IB.
> >
> > - Integrated multi-rail communication support for OpenFabrics
> >   Gen2-iWARP.
> >
> > - Blocking mode of communication progress. Available for OpenFabrics
> >   Gen2-IB.
> >
> > - Based on MPICH2 1.0.5p4.
> >
> > For downloading MVAPICH2 1.0-beta source code, associated user guide
> > and accessing the anonymous SVN, please visit the following URL:
> >
> > http://mvapich.cse.ohio-state.edu
> >
> > All feedbacks, including bug reports and hints for performance tuning,
> > are welcome. Please post it to the mvapich-discuss mailing list.
> >
> > Thanks,
> >
> > MVAPICH Team
> >
> > _______________________________________________
> > mvapich-discuss mailing list
> > mvapich-discuss at cse.ohio-state.edu
> > http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
> >
> 
> 
> -- 
> Eric A. Borisch
> eborisch at ieee.org
> 
> ------=_Part_46971_9804973.1185892355266
> Content-Type: text/html; charset=ISO-8859-1
> Content-Transfer-Encoding: 7bit
> Content-Disposition: inline
> 
> Dr. Panda,<br><br>Thank you again for you and your group&#39;s hard work on this software.<br><br>I&#39;ll start by saying that I know I should move over to OpenFabrics and Gen2, but as we&#39;ve discussed previously, this isn&#39;t currently a viable option for reasons that are outside the scope of this forum. With that said...
> <br><br>A few compilation snags with MVAPICH2-1.0-beta on the VAPI flavor:<br><br>(1) In src/mpid/osu_ch3/channels/mrail/src/rdma/ch3_read_progress.c line 146 :<br><br><div style="margin-left: 40px; font-family: courier new,monospace;">
> type = MPIDI_CH3I_MRAILI_Cq_poll(v_ptr, NULL, 0, is_blocking)<br></div><br>calls with four arguments; the VAPI version ( defined in src/mpid/osu_ch3/channels/mrail/src/vapi/mpidi_ch3_rdma_post.h ) has only the first three arguments. I imagine this is just a missing 
> <span style="font-family: courier new,monospace;">#ifdef</span> switch ...<br><br>(2) in src/mpid/osu_ch3/channels/mrail/src/vapi/rdma_iba_1sc.c lines 151-156 : <br><br style="font-family: courier new,monospace;"><div style="margin-left: 40px;">
> <span style="font-family: courier new,monospace;">if (SMP_INIT)</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">{</span><br style="font-family: courier new,monospace;">
> <span style="font-family: courier new,monospace;">&nbsp;&nbsp;&nbsp; /*correspoding post has not been issued */</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">&nbsp;&nbsp;&nbsp; flag = 0;</span>
> <br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">&nbsp;&nbsp;&nbsp; break;</span><br style="font-family: courier new,monospace;"><span style="font-family: courier new,monospace;">}</span>
> <br></div><br> These lines appear to have migrated here from somewhere else in the code (perhaps&nbsp;&nbsp;the function immediately above it.) The variable <span style="font-family: courier new,monospace;">flag</span> is undefined at this point, and there&#39;s a 
> <span style="font-family: courier new,monospace;">break</span> statement without a loop to break out of...<br><br>By no means a tested fix, but removing the last argument from the issue mentioned in (1) and commenting out the offending lines in (2) appears to allow the VAPI channel to compile and run (benchmarks, in-house tools) successfully. I haven&#39;t been able to get logging working, but that is another discussion.
> <br><br>Your thoughts?<br><br>Thanks again!<br>&nbsp;Eric Borisch<br><br>On 7/26/07, Dhabaleswar Panda &lt;<a href="mailto:panda at cse.ohio-state.edu">panda at cse.ohio-state.edu</a>&gt; wrote:<br>&gt; The MVAPICH team is pleased to announce the availability of
> <br>&gt; MVAPICH2-1.0-beta with the following NEW features:<br>&gt; <br>&gt; - Message coalescing support to enable reduction of per Queue-pair<br>&gt;&nbsp;&nbsp; send queues for reduction in memory requirement on large scale<br>&gt;&nbsp;&nbsp; clusters. This design also increases the small message messaging
> <br>&gt;&nbsp;&nbsp; rate significantly. Available for Open Fabrics Gen2-IB.<br>&gt; <br>&gt; - Hot-Spot Avoidance Mechanism (HSAM) for alleviating<br>&gt;&nbsp;&nbsp; network congestion in large scale clusters. Available for<br>&gt;&nbsp;&nbsp; Open Fabrics Gen2-IB.
> <br>&gt; <br>&gt; - RDMA CM based on-demand connection management for large scale<br>&gt;&nbsp;&nbsp; clusters. Available for OpenFabrics Gen2-IB and Gen2-iWARP.<br>&gt; <br>&gt; - uDAPL on-demand connection management for large scale clusters.
> <br>&gt;&nbsp;&nbsp; Available for uDAPL interface (including Solaris IB implementation).<br>&gt; <br>&gt; - RDMA Read support for increased overlap of computation and<br>&gt;&nbsp;&nbsp; communication. Available for OpenFabrics Gen2-IB and Gen2-iWARP.
> <br>&gt; <br>&gt; - Application-initiated system-level (synchronous) checkpointing in<br>&gt;&nbsp;&nbsp; addition to the user-transparent checkpointing. User application can<br>&gt;&nbsp;&nbsp; now request a whole program checkpoint synchronously with BLCR by
> <br>&gt;&nbsp;&nbsp; calling special functions within the application. Available for<br>&gt;&nbsp;&nbsp; OpenFabrics Gen2-IB.<br>&gt; <br>&gt; - Network-Level fault tolerance with Automatic Path Migration (APM)<br>&gt;&nbsp;&nbsp; for tolerating intermittent network failures over InfiniBand.
> <br>&gt;&nbsp;&nbsp; Available for OpenFabrics Gen2-IB.<br>&gt; <br>&gt; - Integrated multi-rail communication support for OpenFabrics<br>&gt;&nbsp;&nbsp; Gen2-iWARP.<br>&gt; <br>&gt; - Blocking mode of communication progress. Available for OpenFabrics
> <br>&gt;&nbsp;&nbsp; Gen2-IB.<br>&gt; <br>&gt; - Based on MPICH2 1.0.5p4.<br>&gt; <br>&gt; For downloading MVAPICH2 1.0-beta source code, associated user guide<br>&gt; and accessing the anonymous SVN, please visit the following URL:
> <br>&gt; <br>&gt; <a href="http://mvapich.cse.ohio-state.edu">http://mvapich.cse.ohio-state.edu</a><br>&gt; <br>&gt; All feedbacks, including bug reports and hints for performance tuning,<br>&gt; are welcome. Please post it to the mvapich-discuss mailing list.
> <br>&gt; <br>&gt; Thanks,<br>&gt; <br>&gt; MVAPICH Team<br>&gt; <br>&gt; _______________________________________________<br>&gt; mvapich-discuss mailing list<br>&gt; <a href="mailto:mvapich-discuss at cse.ohio-state.edu">mvapich-discuss at cse.ohio-state.edu
> </a><br>&gt; <a href="http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss">http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss</a><br>&gt; <br><br><br>-- <br>Eric A. Borisch<br><a href="mailto:eborisch at ieee.org">
> eborisch at ieee.org</a><br><br>
> 
> ------=_Part_46971_9804973.1185892355266--
> 



More information about the mvapich-discuss mailing list