Your message dated Fri, 12 Aug 2016 09:56:30 +0000
with message-id <[email protected]>
and subject line closing RFP: mcmpi -- Grid-enabled implementation of MPI
has caused the Debian Bug report #528786,
regarding RFP: mcmpi -- Grid-enabled implementation of MPI
to be marked as done.
This means that you claim that the problem has been dealt with.
If this is not the case it is now your responsibility to reopen the
Bug report if necessary, and/or fix the problem forthwith.
(NB: If you are a system administrator and have no idea what this
message is talking about, this may indicate a serious mail system
misconfiguration somewhere. Please contact [email protected]
immediately.)
--
528786: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=528786
Debian Bug Tracking System
Contact [email protected] with problems
--- Begin Message ---
Subject: ITP: mcmpi -- Grid-enabled implementation of MPI
Package: wnpp
Owner: Tooru Fujisawa <[email protected]>
Severity: wishlist
*** Please type your report below this line ***
* Package name : mcmpi
Version : 0.21.0
Upstream Author : Hideo Saito <[email protected]>
* URL : http://www.logos.ic.i.u-tokyo.ac.jp/~h_saito/mcmpi/
* License : GPL
Programming Lang: C, FORTRAN
Description : Grid-enabled implementation of MPI
MC-MPI is a Grid-enabled implementation of MPI, developed by Hideo
Saito at the University of Tokyo. Its main features include the
following:
- [Firewall and NAT traversal]: MC-MPI constructs an overlay
network, allowing nodes behind firewalls and nodes without global
IP addresses to participate in computations. There is no need to
perform maual configuration; MC-MPI automatically probes
connectivity, selects which connections to establish, and performs
routing.
- [Locality-aware connection management]: Establishing too many
connections, especially wide-area connections, results in many
problems, including but not limited to the follwing: exhaustion of
system resources (e.g., file descriptors, memory), high message
reception overhead, and congestion between clusters during
all-to-all communication. Therefore, MC-MPI limits the number of
connections that are established. If we assume, for simplicity,
that n processes are distributed equally among c clusters, then at
most O(log n) connections are established by each process and at
most O(n log c) connections are established between clusters. As
MC-MPI uses a lazy connect strategy, fewer connections are
established for applications in which few process pairs
communicate. The maximum number of connections allowed can be
controlled by passing the -beta option to mpirun (see Subsection 3).
- [Locality-aware rank assignment]: Temporarily disabled in this
version.
--- End Message ---
--- Begin Message ---
RFP 528786 has no visible progress for a long time, so closing.
--- End Message ---
