From glenn@bu.edu  Wed Aug 13 14:35:25 2003
Subject: Going, going, ...
Date: Wed, 13 Aug 2003 14:15:46 -0400
Organization: Boston University

Dear Researcher,

I am very pleased to announce that we are finalizing the installation
of the new IBM p655 and are planning to bring the new system into
production next Tuesday, August 19, at 10:00 AM. This will require
approximately one hour of down-time. While the system changes will
be mostly transparent, there are several issues of which you should be
aware and these are detailed below. Importantly, please note that
the new system is intended to replace the IBM SP, as previously
announced, and that system will be taken out of production at the
end of September.

The new machine is an IBM p655 system with 48 processors configured
as six 8-processor nodes. The nodes are powered by IBM RS6000 Power4
processors running at 1.1GHz with 2GB of main memory per processor.
Since the p655 has the same architecture, runs the same operating
system and layered software products and will be configured similarly
to the other production p690 machines, it should integrate rather
seamlessly with these facilities.

Next Monday at 5:00 PM we will stop dispatching p4-long jobs to
Twister (the login/interactive node). Twister will be brought down
at 10:00 AM on Tuesday for the changes and any jobs that are still
running on it will be terminated. When Twister comes back on-line it
will be one of the new p655 nodes. The 32-processor p690 node that
was formerly Twister will become dedicated to batch, as will the
remaining five new nodes. These machines should all be available by
the end of the day. The table below shows all of the Power4 nodes and
their batch configuration:
        Twister   p655   interactive
        Scrabble  p655   p4-short (2 job slots),
                         p4-verylong (2 job slots),
                         p4-mp4 (1 job slot)
        Marbles   p655   p4-long (8 job slots)
        Crayon    p655   p4-long (8 job slots)
        Litebrite p655   p4-mp8 (1 job slot)
        Hotwheels p655   p4-mp8 (1 job slot)
        Kite      p690   p4-mp16 (2 job slots)
        Frisbee   p690   p4-mp16 (2 job slots)
        Pogo      p690   p4-mp32 (1 job slot)

Please note that in the new configuration several of the batch
queues (p4-long, p4-mp8, p4-mp16) will run on multiple machines.
If you need to reference files on a particular scratch partition,
please make sure you use the fully qualified name (e.g.
/frisbee/scratch/myfile).

Also note that 3 new queues have been added with the following limits:
        Queue         Procs  CPU limit(hrs)   Runtime limit(hrs)
        p4-verylong    1       64              80  
        p4-mp4          4       16               5
        p4-mp8          8       32               5

Finally, we will be imposing wall-clock time limits on all of the single
processor queues, similar to those that exist on all of the mp queues.
The single process "long" queues will have a 40 hour runtime limit, while
the "short" queues will have a 2.5 hour limit. A table showing all the
queues and limits can be found on our Web site at
http://scv.bu.edu/SCV/scf-techsumm.html

Since the new machines run at a slightly slower clock speed than the
p690s (1.1 GHz vs. 1.3 GHz), they will be charged at a correspondingly
lower rate of 4.2 SUs per processor hour (versus 5 SUs per hour for
the p690). Once the SP is retired this fall, we will renormalize all
allocations and charges so that one SU will roughly correspond to one
hour of processor time. 

If you have any questions or concerns, please feel free to contact
me by email (glenn@bu.edu) or phone (617-353-1319). As always, Kadin
Tseng (kadin@bu.edu) or Doug Sondak (sondak@bu.edu) can assist with
any questions regarding programming, software packages or other issues
related to the use of the facilities.

Sincerely,

Glenn Bresnahan