---------------------------------------------------------------------- H H PPP CCC UK NATIONAL HPC SERVICE H H P P C C -------------------------- HHHHH PPP C x x EPCC and CCLRC Daresbury H H P C C xx Laboratory are members of H H P CCC x x the HPC-UK partnership ---------------------------------------------------------------------- HPCx User Mailing 085 16 November 2005 ---------------------------------------------------------------------- Contents ** Migration to Phase 2A ** Phase 2A: what is it like? ** Phase 2A: regions, queues and jobs ** Phase 2A: notes ** Events, November - December 2005 ---------------------------------------------------------------------- Greetings-- MIGRATION TO PHASE 2A As I mentioned in earlier mailings, we are about to move the HPCx service to a new platform, which we refer to as Phase 2A. Unless something really unexpected happens, we plan to close the current Phase 2 service at 10:00 next Wednesday morning, 23 November. Phase 2A will open at 10:00 on Friday morning, 25 November. All the current disks will be transferred to the new platform, so your files will be in the same places as before. The IP addresses of the new service will remain the same, and you will still ssh to login.hpcx.ac.uk. All your passwords and ssh keys will be unchanged. Please note: any jobs that are waiting to run when the Phase 2 service ends will be lost - you will have to submit them again. Apologies once again for the break in service. ---------------------------------------------------------------------- PHASE 2A: WHAT IS IT LIKE? The Phase 2A system is based on the Power 5 processor, instead of the Power 4+ used in Phase 2. The processors are grouped into 16-processor nodes, each of which corresponds to a single LPAR. In all there are 96 such nodes, so that the system has 1536 processors. Each node has 32 Gbytes of main memory, so there is twice as much memory per processor as there was on Phase 2. When run on Phase 2A, the Linpack benchmark shows an Rmax figure of 7.395 Tflops. This compares to 6.188 Tflops for Phase 2, despite the reduction in processor numbers from 1600 to 1536. For interest, eight 16-processor nodes fit in a single cabinet. That's 128 processors per cabinet, in place of the 32 per cabinet of Phase 2. The move to Power 5 technology keeps us in step with IBM's development of this product line, so that we can get the benefit of their software development. It means that when we come to upgrade the service to Phase 3 in the middle of next year, we can simply add more nodes to increase the system's performance. ---------------------------------------------------------------------- PHASE 2A: REGIONS, QUEUES AND JOBS Although the system includes 96 16-processor nodes, for the present we will only be using 92 of them. As before, they will be divided into regions, as follows: 64 nodes (1024 processors) will form the production (or capability) region. 26 nodes (416 processors) will form what will now be called the capacity region - this corresponds to what used to be called the development region 2 nodes (32 processors) will form the interactive parallel region. Jobs requiring more than 128 processors will run in the capability region; jobs of up to 128 processors will run in the capacity region. As the size of the capability region is 1024 processors, it will not be possible to run jobs which are larger than that. The minimum number of processors which will be assigned is 32, or two nodes (as before, this doesn't apply to parallel interactive jobs). If you ask for fewer processors than 32, you will still be charged for 32. As before, for jobs using fewer than 128 processors, the maximum runtime is 6 hours. For jobs of 128 processors and up, the maximum is still 12 hours. A new version of the User Guide will available shortly; also new recommended LoadLeveler scripts as tarballs will be accessible through the FAQ. I shall mail round about this later. ---------------------------------------------------------------------- PHASE 2A: NOTES Existing codes compiled for Power4 will work on the new system. Some people have found improvements in performance by recompiling without changing their compiler options or makefiles. As a mentioned above, there will be about twice as much memory available per processor. Currently, you may be using a "tasks_per_node" line to reduce the number of processors you use, so as to have more memory available per task. If so, you may find that this is no longer necessary. Bear in mind that there are only 16 processors on a node on Phase 2A. The procedure for changing your password on the service machine has been streamlined and made more self-explanatory. There will be a one-day course on optimisation for the Power5 processor at Daresbury Laboratory on Thursday, 8 December. The course will cover the new architecture and optimisation techniques (especially differences from the earlier Power4 architecture), and will also present initial results on the new system. Please register here: http://www.hpcx.ac.uk/support/training/form.html ---------------------------------------------------------------------- EVENTS, NOVEMBER - DECEMBER 2005 HPCx and SC|05 -------------- Come and visit EPCC and CCLRC at SC|05, Seattle, WA, 12-18 November, booths 132/136. Workshop: Progress in Environmental Engineering ----------------------------------------------- Monday, 21 November: Daresbury Laboratory More information: http://www.cse.clrc.ac.uk/ceg/meetings/enveng/ Registration form: http://www.cse.clrc.ac.uk/ceg/meetings/enveng/EnvEng_RegForm.pdf Course: Scientific Visualisation using VTK ------------------------------------------ Wednesday, 23 November: EPCC,Edinburgh More information: http://www.hpcx.ac.uk/support/training/HPCxVis.html Registration: http://www.hpcx.ac.uk/support/training/form.html Course: Parallel IO using MPI-IO -------------------------------- Thursday, 24 November: EPCC, Edinburgh More information: http://www.hpcx.ac.uk/support/training/HPCxMPIIO.html Registration: http://www.hpcx.ac.uk/support/training/form.html Course: Computational Chemistry On HPCx --------------------------------------- Wednesday-Friday, 30 November-2 December: Daresbury Laboratory For further information, please go here: http://www.cse.clrc.ac.uk/events/HPCx_ChemistryTraining/ To register: http://www.cse.clrc.ac.uk/events/HPCx_ChemistryTraining/HPCxCompChem_RegForm.p $ HPCx Annual Seminar: "Capability Science on HPCx" ------------------------------------------------- Monday 5 December: Daresbury Laboratory Further information and registration: http://www.hpcx.ac.uk/about/events/annual2005/ 16th Machine Evaluation Workshop -------------------------------- Tuesday-Wednesday 6-7 December: Daresbury Laboratory Further information: http://www.cse.clrc.ac.uk/disco/mew16/index.html Registration form http://www.cse.clrc.ac.uk/database/MEW16-reg-form.jsp Course: Optimisation For Power5 ------------------------------- Thursday 8 December: Daresbury Laboratory Please register here: http://www.hpcx.ac.uk/support/training/form.html European Global Array Users' Meeting ------------------------------------ Monday-Tuesday 28-29 November: Daresbury Laboratory More information and registration: http://www.cse.clrc.ac.uk/events/GA_UsersMeeting/ ---------------------------------------------------------------------- Regards --John ---------------------------------------------------------------------- HPC-UK partnership: http://www.hpc-uk.ac.uk ---------------------------------------------------------------------- Earlier mailings: http://www.hpcx.ac.uk/support/notices/index.html To be removed from the mailing list: log into your website account, go to the "Update" page, and click the "Opt out of user emails" field; then click "Commit update". -- John Fisher j.fisher@epcc.ed.ac.uk HPCx User Administration and Helpdesk Phone: +44 131 650 5029 HPCx: http://www.hpcx.ac.uk Helpdesk: support@hpcx.ac.uk