On February 1, 1997, NCSA's 512 node CM-5 will be
retired from computational service. The retirement of the CM-5
is the latest step in our long-term strategy to provide scalable,
microprocessor-based, cache-coherent shared memory systems to
our national user community. The CM-5 has been an extremely successful
system at NCSA for large scale, massively parallel simulations,
and we recognize that its retirement is a loss to established
users. NCSA made an effort to extend the time that the CM-5 would
remain in service, but financial constraints have necessitated
the February retirement schedule.
NCSA will be providing assistance in migrating to
the newer platforms located at the Center. These include the SGI
POWER CHALLENGEarray and HP/Convex EXEMPLAR SPP-1200, both in
production use. In addition, in October, NCSA installed recently
announced SGI Cray Origin2000 and HP/Convex EXEMPLAR S Class systems.
The HP/Convex EXEMPLAR SPP-1200 and S Class systems are both distributed shared memory architecture machines. The SPP-1200 currently consists of 4 HYPERnodes of 8 HP PA-RISC 7200 processors and 1 Gbyte memory each. The EXEMPLAR S Class system is currently a single HYPERnode with 16 HP PA-RISC 8000 processors and 4Gbytes of memory. The S Class systems will be upgraded to an X Class systems with 64 processors and 16Gbytes of memory in early 1997. Further details on the hardware and software configurations of these systems can be found at:
http://www.ncsa.uiuc.edu/General/Consulting/HPCC/Systems/exemplar/
http://www.ncsa.uiuc.edu/General/Consulting/HPCC/Systems/SClass/
The SGI POWER CHALLENGEarray is a network of symmetric multi-processor (SMP) systems and currently consists of five R10000-based POWER CHALLENGEs (four with 16 processors and one with 8 processors) and three R8000-based POWER CHALLENGEs (two with 18 processors and one with 16 processors). The SGI Cray Origin2000 system is a distributed shared memory architecture and currently consists of 128 R10000 processors and 24 Gbytes of memory. The Origin2000 will be upgraded to 256 R10000 processors and 64Gbytes of memory very early in 1997. Further details on the hardware and software configurations of these systems can be found at:
http://www.ncsa.uiuc.edu/General/Consulting/HPCC/Systems/SGI/
http://www.ncsa.uiuc.edu/General/Consulting/HPCC/Systems/Origin2000/
Both the Origin2000 and EXEMPLAR S/X Class systems will be made available in a Friendly User mode as these new systems stabilize. It is expected that initial access will be available in December 1996. Usage on these systems during the Friendly User period will not be charged. These systems will remain in Friendly User mode until such time as they are deemed ready for production use. The proposed timeline calls for production use in the second or third calendar quarters of 1997. Allocation for these systems will be available through the normal processes, but large users are reminded to make requests of > 40,000 SUs through the MetaCenter Allocations process. These new systems are expected to become production computational platforms during the next allocation period. Information regarding the Allocation process is available at:
http://www.ncsa.uiuc.edu/General/Allocations/ApplyTop.html
NCSA will conduct CM-5 Migration Workshops training
sessions during the transition period. The first of these is scheduled
December 9-10. Another workshop will be held in mid-January. A
subsequent notice will be sent when the dates are arranged. Registration
for these may be emailed to
hpc-training@ncsa.uiuc.edu.
Any questions
of a technical nature may be directed to the National Consulting
Office, by email to
consult@ncsa.uiuc.edu
or by phone 217-244-1144.
Comments about the transition, upgrade strategy, or schedule may
be directed to future@ncsa.uiuc.edu
or to me.
To make this transition as smooth as possible, we
are developing a team consisting of NCSA consultants and application
programmers, vendor software experts, and experienced users. This
team will produce training materials and provide assistance with
code migration and testing through training and by providing additional
support to the Consulting Office staff. In addition, we need your
assistance during the transition period. Please make an effort
to get on the alternative machines early, run your codes, and
let us know as soon as possible about any problems you might experience.
This will allow us to resolve any problems or bottlenecks as quickly
as possible, and to build a knowledge base that will be of benefit
to all users.
Thank you,
NCSA Director