Report on the Fourteenth Meeting of Computing Representatives, 27 – 28 May 2002

Report on the Fourteenth Meeting of Computing Representatives, 27 – 28 May 2002

376 Report on the fourteenth meeting of Computing Representatives, 27 – 28 May 2002 P. Prior (Compiler) Operations Department August 2002 Report on the fourteenth meeting of Computing Representatives, 27-28 May 2002 Contents Preface . .2 Part I: ECMWF Staff contributions ECMWF’s computing service: Status and Plans – W. Zwieflhofer . .4 IBM HPCF – N. Storer . .8 Migration to the IBM HPCF – M. Hamrud . .10 ECaccess – a portal to access ECMWF – L. Gougeon . .12 Computing Representatives: Their rôle and responsibilities – U. Modigliani . .14 User registration – U. Modigliani . .19 Graphics update – J. Daabeck . .20 MARS and dissemination update – J. Hennessy . .22 Web development – A. Brady . .24 Part II: Member State and Co-operating State contributions Austria . 28 Belgium . 37 Croatia . .40 Denmark . 42 Finland . 45 France . 47 Germany . 51 Greece . 55 Hungary . .56 Iceland . 59 Ireland . 61 Italy . .67 Netherlands . 69 Norway . 74 Spain . 78 Sweden . 83 Switzerland . 85 United Kingdom . 87 EUMETSAT . .90 Annex 1 – Participants . 94 Annex 2 – Programme . 95 Technical Memorandum No. 376 Report on the fourteenth meeting of Computing Representatives, 27-28 May 2002 Preface The fourteenth meeting of Computing Representatives took place on 27-28 May 2002 at ECMWF. Eighteen Member States and Co-operating States, plus EUMETSAT, were represented. The list of attendees is given in Annex 1. The Head of the Computer Division (Walter Zwieflhofer) opened the meeting and welcomed representatives. He gave a presentation on the current status of ECMWF’s computer service and plans for its development. Each Computing Representative then gave a short presentation on their service and the use their staff make of ECMWF’s computer facilities. There were also presentations from ECMWF staff members on various specific developments in the ECMWF systems. The full programme is given in Annex 2. This report summarises each presentation, with additional detail on those topics expected to be of particular interest to Member States and Co-operating States. Part I contains ECMWF’s contributions and general discussions. Part II contains Member States’ and Co-operating States’ contributions; all the reports were provided by the representa- tives themselves. 2 Technical Memorandum No. 376 Report on the fourteenth meeting of Computing Representatives, 27-28 May 2002 Part I ECMWF Staff contributions Technical Memorandum No. 376 3 Report on the fourteenth meeting of Computing Representatives, 27-28 May 2002 ECMWF Computing Service: Status and Plans – W. Zwieflhofer The configuration of ECMWF’s computer system current at May 2002 is given in Fig. 1 below. FUJITSU VPP5000/100 FUJITSU VPP700/116 VPP 5000 ADIC IBM RS/6000 FUJITSU VPP700E/48 DISASTER RECOVERY SYSTEM HIPPI STK silos PCs (LINUX) SGI Origin 2000 FDDI SGI IBM SP IBM SP OFFICE AREA Firewall IBM P690 IBM P660 H50 HP K580 RMDCN INTERNET Web server IBM RS/6000 HP L1000 Robot tape libraries MEMBER STATES and GIGABIT ETHERNET CO-OPERATING STATES WIDE AREA NETWORK COMPUTER HALL Figure 1 Fujitsu service The Fujitsu machines have been very stable and provided an extremely reliable service over the past year; average user availability has been above 99%. Average CPU utilisation has been approximately 85%, slightly lower than in the previous year because of the increasing complexity and size of some of the jobs submitted. The average percentage of the available processing power used by Member States’ users has increased and is now close to a weekly average of 15%. Fujitsu replacement The procurement process took place over most of 2001. There were four replies to the Invitation To Tender (Cray, Fujitsu, IBM and NEC). The benchmark used was complex and comprehensive, comprising a full 4D-Var data assim- ilation at fairly high resolution; a deterministic forecast at a resolution higher than the current operational version and a 50-member EPS. Vendors were required to commit contractually to their equipment’s performance on each of the three elements of the benchmark. Following thorough tender evaluation over four months, IBM was selected as the supplier of the next High Performance Computing Facility (HPCF). The equipment will be rented and the contract will expire on 31 March 2007. The IBM High Performance Computing Facility The basic building blocks of the system are IBM p690 servers, each of which has 32 processors. Each server will be divided into four 8-processor partitions (nodes), using the Logical Partitioning (LPAR) feature; since each node has its own switch cards, the switch capacity of the whole system can be increased in this way. Each partition runs its own copy of the operating system, AIX. The parallel file system that will be provided GPFS (Global Parallel File System), 4 Technical Memorandum No. 376 Report on the fourteenth meeting of Computing Representatives, 27-28 May 2002 is rich in features and options; batch job submission and execution will have to migrate from NQS to LoadLeveler; there will be a need for a period of familiarisation before their advantages can be fully exploited. The system will be divided into two identical clusters. System maintenance can be performed on the clusters separately and individual nodes can be taken out of service and reintroduced without affecting the operation of the cluster. It is anticipated that these features will enhance system stability. More information is given in a later presentation. Table 1, below, lists the characteristics of the system to be delivered at each of the three phases. Each of these phases must meet a contractually guaranteed level of performances (1). PHASE 1 PHASE 2 PHASE 3 Processor 14.3 GHz Power4 14.3 GHz Power Faster Power Number of processors 10408 1092 ~300 Interconnect D)ual-Colony (PCI) Dnual-Colony (PCI-X Federatio I/O Nodes 82NH-2 8nNH- 4 p690 follow-o Disk space (Fibre Channel) 8B.4 TB 1B2.4 T 27 T Sustained performance 1s.3 x VPPs 1s.9 x VPP 5.2 x VPP Access to the IBM HPCF for Member State users One Phase 1 cluster will be delivered to ECMWF in June/July 2002 to enable the development of an ECMWF specific environment and the fine-tuning and adjustment of the file system and queuing structures. Member State users who have a particular need to gain early experience with the new system will be given the opportunity to access a ‘famil- iarisation configuration’ in its immature state from 1 August; this possibility will not be generally announced but Computing Representatives are invited to offer the opportunity to those users they think would benefit and these users may discuss their needs with their User Support Contact Point. It is planned that an official trial service for Member State users will start at the beginning of November 2002. This system will be suitable for the migration of Member State codes and scripts, although file system changes may be required and there may be more frequent interruptions of service than usual for reconfigurations and similar. Provided that the 30-day Operational Acceptance is passed by 23 December 2002, it is planned to provide a full user service from January 2003. The IBM and Fujitsu systems will then run in parallel until the end of March 2002, when the Fujitsu service will be ended. (Failure of the IBM system to pass Operational Acceptance by 23 December will necessitate the extension of the Fujitsu contract beyond March 2003.) Migration of Member States’ workload The contract with IBM includes twenty-four man-months of support (advice and general problem solving) from IBM applications specialists; these resources will be shared with the Member States and should be applied for via the appropriate User Support Contact Point. GPFS is very different from the current Fujitsu I/O subsystem. ECMWF will formulate advice, once some experi- ence has been gained in-house. Training There are three training courses planned in 2002: 24 – 28 June How to use an IBM cluster 1600 (including AIX, LoadLeveler, compilers, etc.). An ECMWF specific environment will not be available; a single p690 only will be available. 28 Oct – 1 Nov How to use an IBM cluster 1600, (including AIX, LoadLeveler, compilers, etc.). and 18 – 22 Nov An ECMWF specific environment will be available; practical training will use the full cluster. (1) In the period between the Computer Representatives’ Meeting and the publishing of this report, tests have revealed that more processors than shown in Table 1 will be required to meet the contractual performance levels. Technical Memorandum No. 376 5 Report on the fourteenth meeting of Computing Representatives, 27-28 May 2002 Data Handling System Following a procurement in 2002, it was decided to replace the existing TSM-based (Tivoli Storage Manager) Data Handling System with a new system based on HPSS (High Performance Storage System). HPSS is a hierarchical.

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    98 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us