Welcome at SARA

1
Welcome

• Welcome at SARA!
• Willem Vermin
• Senior consultant
• User support
• Optimization and parallelization

2
SARA who we are

• Founded in 1971 as computing centre for the
  Universities of Amsterdam and the Mathematical
  Centre.
• Now supplier of advanced services in the domain
  of networking, supercomputing, virtual reality,
  housing etc. for the academic community in the
  Netherlands, but also for commercial users.
• Important as national supercomputer centre.
• About 70 FTE

3
SARA services

• High Performance Networking
• Management of Surfnet
• Visualization and Information services
• CAVE facility
• System Services
• CIO (Computers in Education)
• Housing
• High Performance Computing

4
SARA Focusing on HPC

• Hosting National Supercomputer for Academic
  Netherlands (Now 1 Teraflop, soon 3.2 Teraflop)
• Support using the facilities
• Optimizing and parallelization of existing user
  codes
• Grid computing
• Hosting (super) computers for other institutes
• University of Amsterdam (Beowulf cluster)
• Free University of Amsterdam and FOM research
  institutes (IBM Cluster1600)
• Backup services (TSM)

5
Focus on National Supercomputer

• Users from all universities in the Netherlands
• Simulation (currents, gases, )
• Biochemical calculations (DNA,)
• Weather climate trends
• Chemical calculations
• About 500 active users
• Paid by government
• Commercial use

6
Focus How to use supercomputer

• Operating system Unix
• Access via Internet
• Parallel system
• Adaptation of user code necessary
• SARA assists in
• Optimization programming style and optimized
  libraries
• Parallelization of codes
• Speedup 10-1000

7
Focus Grid technology

• Computing power and storage facilities
• As easy as electrical power in your home
• User doesnt care or know where jobs are run or
  data is stored
• Requires international collaboration
• SARA
• European Datagrid
• Dutchgrid

8
Focus end of presentation

• Questions?
• willem_at_sara.nl
• http//www.sara.nl

9
SGI Origin 3800

• Theoretical peak performance 1 Tflop/s
• 1024 x 500 Mhz R14000 processors
• 1 TB memory
• 10 TB disk

10
Storage and back-up facilities

• StorageTek
• 9840 tape drives (10) - 20 GB ? 40 GB
• 2500 tapes 50 TB
• 9940 tape drives (2) - 200GB
• 750 tapes 150 TB
• Hierarchical data management (DMF, TSM)
• Automatic backup facilities (TSM)
• IBM Magstar tape robot (3494)

11
Cluster1600 configuratie

• One frame met 3 winterhawk nodes
• 4 power3 processors (375 MHz)
• 4 GB memory per node
• Gigabit ethernet
• 4 p690 nodes
• 32 power4 processors (1.1 GHz)
• 32 GB memory per node
• Gigabit ethernet
• PCI-based SPswitch2 adapter
• SP switch (250 MB/s)
• GPFS File system (0.5 TB)

12
Cluster1600 Huidige configuratie
13
Beowulf Linux clusters

• UvA
• 126 AMD Athlon 700 MHz nodes
• 42 AMD Athlon 1200 MHz nodes
• disk 8 Gbyte/node
• network 100 Mbit switched ethernet between
  nodes, 1 Gbit ethernet to the fileservers
• 3 Intel dual CPU file servers
• 200 GB IDE RAID per server
• 9 AMD Athlon 1600XP 1400 MHz Testcluster
• Gigabit ethernet
• 2 Itanium I nodes (4 CPUs) 700 MHz
• 2 36GB disks
• UvA/SARA

14
Beowulf management

• Debian 3.1
• Dpkg tools
• Lokale mirror
• Tools
• Systemimager
• Pxe compatibele netwerkkaarten
• Cfengine
• Distributie van systeem files
• Monitoring van o.a. filesystemen en processen
• Eigen ontwikkeling
• Scheduling
• OpenPBS