SGI Challenges at D0
What is an SGI Challenge?
In the first approximation it is a multiprocessor system based on
MIPS R4400MC 64-bit RISC CPUs with very high
I/O bandwidth, capable of delivering data in excess of 320 Mbytes/second.
We have two Challenges, d0cha and d0chb, for D0 users.
They are both Challenge XL systems.
d0cha is used mainly as
PIAF server. It has twelve processors each
running at 150MHz and 1536 Mbytes of shared memory expandable up to 6
Gbytes.
d0chb is used mainly as interactive/small batch system.
It has eight 200MHz CPUs and 768 Mbytes of shared memory.
Each processor in all three Challenges has
16 Kbytes on-chip instruction cache, 16 Kbytes on-chip data cache, and
1 Mbyte of a secondary cache. The processors inside each Challenge system can
communicate amongst themselves over a high performance system bus with a
bandwidth of 1.2 Gbyte/sec.
For each Challenge, High-Speed I/O system can deliver data rate up
to 320 Mbytes/sec. In addition there is a VME64 bus with a bandwidth of
50 Mbytes/sec.
Each system can handle up to 24 SCSI-2 channels (with up to 15 devices per
channel) with a maximum bandwidth of 20 Mbytes/sec per channel.
Why Challenge?
Challenge is an example of a class of strongly coupled computer systems:
it consists of an expandable collection of high performance processors,
accessing a very large data set locally, i.e. without any local
area network. Other systems of this type include: DEC Sables, Convex Exemplar,
IBM SP1/SP2 systems. The SGI Chalenge was selected because of its high
performance/price ratio. An additional argument that our D0 software is
already ported to the SGI platform was of importance too.
What for?
This is our first excursion in the area of more centralized, UNIX based
central computing systems. Its use will probably evolve with time and our
experience. We envisage a scenariom where all of the D0 data will migrate
to such a system, be backed up by high capacity robotic mass storage
under a convenient Hierarchichal Mass Storage Management System.
We plan to evaluate its use to perform the following functions :
- Central Analysis Engine: taking advantage of very high I/O bandwidth all
jobs requiring access to a very large data sets should run efficiently there
- UNIX-based File Server (or Event Server, or Object Server) delivering
smaller data sets to the analysis clusters located at D0.
- Unexpected: if the system performs anywhere near expectations, many of
us will find multiple uses of it for our favoured physics projects.
Looking forward....
Disks Available
d0cha
Data disks
| disks |
size(GB) |
owner |
| /proj1 |
6.5 |
QCD |
| /proj2 |
6.5 |
QCD |
| /proj3 |
6.5 |
QCD |
| /proj4 |
6.5 |
QCD |
| /proj5 |
6.5 |
WZ |
| /proj6 |
6.5 |
WZ |
| /proj7 |
6.5 |
WZ |
| /proj8 |
6.5 |
WZ |
| /proj9 |
6.5 |
TOP_LJETS |
| /proj10 |
6.5 |
TOP_LJETS |
| /proj11 |
6.5 |
TOP_LJETS |
| /proj12 |
6.5 |
TOP_LJETS |
| /proj13 |
6.5 |
TOP_ALLJETS |
| /proj14 |
6.5 |
TOP_ALLJETS |
| /proj15 |
7.7 |
QCD_ESCALE |
| /projects/707 |
6.5 |
QCD |
| /projects/708 |
2.2 |
TAU |
| /projects/709 |
8.0 |
WZ |
| /projects/710 |
8.0 |
TOP |
| /projects/712 |
2.2 |
UPG_TRG |
| /projects/713 |
2.2 |
SVX |
| /projects/714 |
8.9 |
NEW |
| /projects/715 |
8.9 |
QCD |
| /projects/716 |
8.9 |
TOP |
| /input |
6.5 |
general |
| /output |
6.5 |
general |
|
Spool/Scratch disks
| disks |
size(GB) |
owner |
| /spool1 |
2.7 |
general |
| /spool2 |
6.5 |
general |
| /spool3 |
6.5 |
general |
| /spool4 |
6.5 |
general |
| /spool5 |
6.5 |
general |
| /spool6 |
6.5 |
general |
| /spool7 |
4.3 |
WZ |
|
d0chb
Data disks
| disks |
size(GB) |
owner |
| /projects/703 |
8.9 |
unknown |
| /projects/704 |
8.8 |
B_PHY |
| /projects/705 |
8.8 |
QCD |
| /projects/706 |
8.9 |
WZ |
| /projects/711 |
2.2 |
ONLINE |
|
Spool/Scratch disks
| disks |
size(GB) |
owner |
| /spool0 |
13.3 |
general |
| /scratch3 |
6.6 |
general |
| /scratch5 |
4.4 |
general |
|
last updated 5/19/97 by Dong Zhao