next up previous
Next: Grid computations using private Up: latex2html_globus Previous: Introduction

MecaGRID resources

The computer resources for the MecaGRID are the clusters at the different members of the MecaGRID to be combined to function as one large parallel computer giving MecaGRID users the capability to perform computations that are not possible at any member site (the maximum processors available at any member site in the present MecaGRID configuration is 70 at INRIA Sophia Antipolis).


Table 1: Total MecaGRID processors (public + private)
Site cluster IPA CPUs GHz RAM/Node LAN p/N  
INRIA Alpes icluster1 public 216 0.7 3.00 GB 0.100 1  
INRIA Alpes icluster2 public 200 1.0 3.00 GB 1.000 2  
INRIA sophia INRIA-pf public 38 1.0 0.50 GB 0.100 2  
INRIA sophia INRIA-nina public 32 2.0 1.00 GB 1.000 2  
IUSTI m3h-cluster private 30 2.0 1.00 GB 0.100 1  
CEMEF sarek-cluster private 63 1.0 0.50 GB 0.100 2  
Total CPUs   pub+pri 679          


Clusters have either public or private IP addresses. Table 1 shows that 679 processors would be available if both private and public IP addresses were included. The local area network (LAN) speed is in Giga bits per second (Gbps). p/N is the number of processors per Node.

At the present time, users at the individual member sites may experience long wait times when more than 8-16 processors are requested. Depending on the work load, the wait times may extend from several hours to a week. One could imagine that with 679 processors available, computations requiring 32-64 processors (less than 10 percent of the available processors) would not require long wait periods using the MecaGRID. A recent example experienced by the author was a request for 48 processors on the INRIA cluster to evaluate performance that remained in the queue three days before obtaining the 48 processors. The same run was submitted to the MecaGRID where 48 processors were immediately available permitting performance evaluation of the inter-clusters used in the run.

In the summer 2003 INRIA the Rhone-Alpes cluster underwent an upgrade lasting several months and became an inactive member of the MecaGRID. This upgrade is now complete and it is in the interest of the MecaGRID that INRIA Rhone-Alpes be integrated as an active member as its two clusters have three times as many available processors than the other three members combined!



Subsections
next up previous
Next: Grid computations using private Up: latex2html_globus Previous: Introduction
Stephen Wornom 2004-09-10