|
|
|
|
(This page is best viewed with CSS style sheets enabled)
» Asim
» mpc
» mpc.lip6.fr
 |
ASIM team
LIP6 Laboratory
Paris, France |
 |
NEW (feb 2002) : the new MPC/MPI cluster of workstations :
16 proc. 1GHz, 8 Gb RAM, 1/3 To disk - Now available online !
This new cluster is specifically designed to run MPI jobs in a high parallel environment.
You can now launch
your MPI jobs on our MPC machine. You can find more informations
about this here (in french).
historical : the first MPC Parallel Machine that was available online
The MPC
Parallel Computer of LIP6 laboratory was available online through a Job Management System.
You can now launch
your PVM jobs on our MPC machine. You can find more informations
about this here (in french).
|
| What is MPC ?
MPC is a parallel computer. Our project goal is to design a high
performance/low cost multiprocessor, using standard Pentium based
PC-boards as nodes, Unix FreeBSD operating system, and a high
speed communication network built on the gigabit-HSL (High Speed Link)
IEEE-1355 technology.
|
Performances
PUT low level access driver (remote-write by means of DMA accross the network) :
SLR/V (virtual communication channels) :
These results include the user-level call to this kernel protocol.
PVM :
- TF/TF
communications entre deux tâches filles distantes
- TF/TF locales : communications entre deux tâches filles
locales à un même noeud
- TM/TF : communications entre la tâche maître et une
tâche fille (via le démon et donc via Socket Unix)
Software installation
Click here to see some major informations
about the software installation process.
Full description of MPC-OS (software components of MPC)
Click here to get the Ph.D Thesis manuscript of A. Fenyo : it contains a full description of the major software components relative to MPC.
Hardware Components
The high speed network is based on the
RCube dynamic router, developed at UPMC/LIP6. RCube is a VLSI
circuit implementing a 8x8 crossbar with interval labeling and
wormhole routing. Each processor node is connected to a router by means of
another dedicated chip, PCIDDC, implementing a Direct Deposit
Protocol to provide inter-node DMA throw the PCI bus.
A new programmable PCI interface chip, NOE, is under development.
|
Software Components
A stack of kernel protocols and
monitoring daemons has been developed in cooperation with PRiSM and ENST
laboratories, and integrated inside Unix FreeBSD to provide several
communication layers for developpers. An optimized implementation of PVM
on top of this stack is now available.
|
Web servers relative to MPC
|
NEWS !
- A video clip that introduces MPC is now available in MPEG-1 format !
|
|
|
|
![[PRINT]](/ssi/print.gif "Afficher ce document dans une nouvelle fenetre pret pour l'impression"){kind=small}
