From siavash.ghiyasvand at [hidden] Wed Jun 30 14:56:57 2010 From: siavash.ghiyasvand at [hidden] (siavash ghiasvand) Date: Thu, 1 Jul 2010 00:26:57 +0430 Subject: [Mpi-22] reConfigurable MPI Message-ID: Hi, In MPI22 Standard document [*1*] on page 305 lines 27-30 we can see this paragraph: > The MPI Forum decided not to address resource control because it was not > able to > design a portable interface that would be appropriate for the broad spectrum > of existing > and potential resource and process controllers. Resource control can > encompass a wide > range of abilities, including adding and deleting nodes from a virtual > parallel machine, ... > Which means, an MPI-based cluster is not reconfigurable at run time! but in EuroPVM/MPI 2006 a reconfigurable version of MPI for HPC clusters was described by University of New Mexico Scalable Systems Lab [*2*] and they show that their implementation will enhance performance (the main goal of MPI). So, why in MPI22 standard "*The MPI forum decided not to address resource control*" which can improve both flexibility and performance of an MPI-based cluster? Ref: MPI/CTP: A Reconfigurable MPI for HPC Applications Regards, Siavash * -------------- next part -------------- An HTML attachment was scrubbed... URL: From david.solt at [hidden] Wed Jun 30 15:07:53 2010 From: david.solt at [hidden] (Solt, David George) Date: Wed, 30 Jun 2010 20:07:53 +0000 Subject: [Mpi-22] reConfigurable MPI In-Reply-To: Message-ID: Hi Siavash, A couple of reasons for this statement: 1) The 2.2 standard was an update to the 2.0 standard which predates the 2006 paper you referenced. 2.2 only introduced new functions where there was an urgent need. New features are intended for the forthcoming 3.0. 2) The resource control paragraph is in the context of process creation and my quick 1 minute scan of the paper you referenced indicates that it is discussing reconfiguring communication and does not talk about dynamic processes. Thanks, Dave Solt From: mpi-22-bounces_at_[hidden] [mailto:mpi-22-bounces_at_[hidden]] On Behalf Of siavash ghiasvand Sent: Wednesday, June 30, 2010 2:57 PM To: mpi-22_at_[hidden] Subject: [Mpi-22] reConfigurable MPI Hi, In MPI22 Standard document [1] on page 305 lines 27-30 we can see this paragraph: The MPI Forum decided not to address resource control because it was not able to design a portable interface that would be appropriate for the broad spectrum of existing and potential resource and process controllers. Resource control can encompass a wide range of abilities, including adding and deleting nodes from a virtual parallel machine, ... Which means, an MPI-based cluster is not reconfigurable at run time! but in EuroPVM/MPI 2006 a reconfigurable version of MPI for HPC clusters was described by University of New Mexico Scalable Systems Lab [2] and they show that their implementation will enhance performance (the main goal of MPI). So, why in MPI22 standard "The MPI forum decided not to address resource control" which can improve both flexibility and performance of an MPI-based cluster? Ref: MPI/CTP: A Reconfigurable MPI for HPC Applications Regards, Siavash * -------------- next part -------------- An HTML attachment was scrubbed... URL: From wgropp at [hidden] Wed Jun 30 15:12:08 2010 From: wgropp at [hidden] (William Gropp) Date: Wed, 30 Jun 2010 15:12:08 -0500 Subject: [Mpi-22] reConfigurable MPI In-Reply-To: Message-ID: <79FA4987-0415-4976-BB7C-FEC9F09B1CF4@illinois.edu> I think the statement explains it - during the MPI 2 discussions, there was a lot of effort spent on defining a portable, standard, resource control specification. The forum was unable to reach a consensus, and rather than make a poor specification, they left it undefined. This permits MPI implementations to explore interfaces with their resource management systems (and permits an MPI-based cluster to be reconfigurable), just not in a MPI-standard defined way. Note that the info argument to the spawn routines, for example, provides a way to communicate with a resource manager; it was an explicit goal of the design of info to provide this way to communication information that is outside of the specification. Bill On Jun 30, 2010, at 2:56 PM, siavash ghiasvand wrote: > Hi, > > In MPI22 Standard document [1] on page 305 lines 27-30 we can see > this paragraph: > > The MPI Forum decided not to address resource control because it was > not able to > design a portable interface that would be appropriate for the broad > spectrum of existing > and potential resource and process controllers. Resource control can > encompass a wide > range of abilities, including adding and deleting nodes from a > virtual parallel machine, ... > > Which means, an MPI-based cluster is not reconfigurable at run time! > but in EuroPVM/MPI 2006 a reconfigurable version of MPI for HPC > clusters was described by University of New Mexico Scalable Systems > Lab [2] and they show that their implementation will enhance > performance (the main goal of MPI). So, why in MPI22 standard "The > MPI forum decided not to address resource control" which can improve > both flexibility and performance of an MPI-based cluster? > > > Ref: > MPI/CTP: A Reconfigurable MPI for HPC Applications > > > Regards, > Siavash > William Gropp Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign * -------------- next part -------------- An HTML attachment was scrubbed... URL: From siavash.ghiyasvand at [hidden] Wed Jun 30 15:58:19 2010 From: siavash.ghiyasvand at [hidden] (siavash ghiasvand) Date: Thu, 1 Jul 2010 01:28:19 +0430 Subject: [Mpi-22] reConfigurable MPI In-Reply-To: Message-ID: > > 1) The 2.2 standard was an update to the 2.0 standard which predates > the 2006 paper you referenced. 2.2 only introduced new functions where > there was an urgent need. New features are intended for the forthcoming > 3.0. > The forum was unable to reach a consensus, and rather than make a poor > specification, they left it undefined. Now I completely got the "Why?" answer. Thanks. > 2) The resource control paragraph is in the context of process > creation and my quick 1 minute scan of the paper you referenced indicates > that it is discussing reconfiguring communication and does not talk about > dynamic processes. > Note that the info argument to the spawn routines, for example, provides a > way to communicate with a resource manager; it was an explicit goal of the > design of info to provide this way to communication information that is > outside of the specification As you said, That paragraph is in the context of process creating and yes! my question was about runtime reconfiguration of an MPI cluster (whole) based on dynamic application requirements. But I think that would be impossible without having a flexible resource control. When a new node joins a cluster some pre-works must be done to enable that new node act as a member of that cluster (in run time), also, some one must be informed about new resources which just have been added to the cluster by entrance of new node and I think all these preparations need that missing flexible resource manager. Regards, Siavash * -------------- next part -------------- An HTML attachment was scrubbed... URL: