|
COMPUTING
|
|
High Performance Computing
In the last UGAMP newsletter I described UGAMP's current High Performance Computing (HPC) allocation. We reached a crisis point in the Summer of 1995 and after negotiation with RAL and NERC we arrived at an arrange-ment which enabled UGAMP to continue using the YMP8 but at a much reduced level. We have been asked to moderate our usage to ensure that other users get their fair share of this over allocated, heavily used facility. This has worked well so far but our usage has increased over the last few weeks. We have a very small amount of our current allocation left so it is still important that we continue to moderate our use of the YMP8 for another few weeks until it is replaced by the J90.
The first piece of good news is that the YMP8 will shortly be replaced by a 32 processor Cray J90. This new system is already in place and working. Hopefully early user trials will begin shortly while the acceptance tests are run. UGAMP are planning to take part in the early user trials. The down side of this news is that a J90 processor is 0.6 of the speed of a YMP processor, but as there are 32 processors we should see a great improvement in our job turnaround. I have heard that RAL are planning to double allocations on the new machine which would mean that from April 1996 we can hope for an allocation of 32000 CPU hours per year.
The second piece of good news is that the Science Budget allocations, announced by the Of- fice of Science and Technology, have a provisional figure of 10M pounds for a new national HPC procurement. This means that the procurement procedure can now begin with the hope that the new machine(s) can be in place at the beginning of 1997. The J90 at RAL was announced as an interim measure to alleviate the pressure on the national vector facilities. At the same time as this vector upgrade, the Cray T3D at Edinburgh has been upgraded to 512 processors which makes it the largest T3D outside the US. So clearly the new procurement will involve both scalar, vector and parallel national facilities. UGAMP via NERC will be involved in this procurement exercise so if you have any thoughts about this new procurement then please contact me.
Lois Steenman-Clark (Reading, CGAM)
T3D News
UGAMP was successful in its bids to both EPSRC and NERC for a scientific support post under the High Performance Computing Initiative. Ben Edgington and Cate Bridgeman took up these posts at Reading and Cambridge respectively in October 1995. They will benefit from the pioneering work on the T3D by Jeff Cole and Nicholas Pinhey to begin to port more of our UGAMP models to massively parallel systems.
Nicholas Pinhey hopes to run David Tan's experiment, which was described at the September UGAMP meeting, on all 512 processors when the machine upgrade has been completed, to achieve an even higher resolution.
Jeff Cole is completing the runs on the T3D for the current round of SGCM experiments and is preparing the parallel and portable versions of the code for release when the documentation and testing are complete.
Cate Bridgeman is working on porting SLIMCAT to the T3D, as the ACSMU chemistry packages are ideal candidates for parallelisation. Cate is implementing the ECMWF semi-lagrangian scheme in SLIMCAT and Ben Edgington has written the communications part of this scheme for the T3D using strategies developed at ECMWF. This experience with the parallel semi-lagrangian scheme should be useful when we try to implement the parallel version of the IFS. As input/output is such a crucial issue on the T3D and as SLIMCAT has never used any packing to reduce its output Ben has GRIB encoded SLIMCAT/TOMCAT output and is providing various GRIB decoding utilities. Details will be announced shortly when testing has been completed.
Plans for porting IFS, UM and coupled modelling work are also being developed.
Lois Steenman-Clark (Reading, CGAM)
Software News
(see also Software and Support)
New tools, either written within UGAMP or made available by other sites, are often found in many different forms or versions throughout UGAMP. As these tools are usually not well publicised it is difficult to find out sufficient information about them, hence the temptation to develop yet another tool! In an attempt to gather together information about software tools used within UGAMP I thought that it would be useful to use the UGAMP Newsletter to share information about useful tools. As a start I have listed below some new or new versions of some software tools. For further information on these tools or if you have any tools you would like to see better publicised for UGAMP use then please contact me.
ugrib
This was written by Jeff Cole for ECMWF GRIB files to extract a specified date or print information about GRIB records or to convert blocked files to unblocked files or vice-versa. The tool works on the Cray YMP8. The executable is in /home/ymp8/kd/bin/ugrib and the documentation is in /home/ymp8/kd/doc/ugrib.README.
xgrib
This was written by John Chambers at ECMWF and it is a X interface to a similar tool to ugrib. Jeff Cole has installed it on our Reading Suns and ported the tool to the Cray. Warning, it is a bit slow on the YMP8 as the interactive response is poor, this should be better on the J90. The executable is in /home/ymp8/kd/bin/xgrib and there is a rudimentary on-line help system.
IFS13r4 source code browser
The script which builds hypertext images of fortran programs was developed by Meteo France. The script we have been given by ECMWF can be used for programs that adhere to ECMWF naming conventions and Jeff Cole has applied this script to the IFS version 13r4 source code. The IFS13r4 code browser has been made available via the Reading UGAMP Internal web page.
libFutf.a
This library of low level routines for reading and writing UTFÕs both version 1.3 and 1.4 was written by Conrad Winchester and tested and debugged by Jason Lander. Version 2.3 has now been installed in the directory /home/ymp8/kd/lib.
Lois Steenman-Clark (Reading, CGAM)