Blizzard: modeling the climate change

Impressive and elegant new German supercomputer:



Specifications:

HPC Cluster ("blizzard")

  • IBM p575 "Power6" cluster
  • water cooled
  • 16 dual core CPUs per node
  • Total size: 264 nodes, 8448 cores
    • 249 compute nodes
    • 12 I/O nodes
    • 3 interactive nodes
  • 18,8 GigaFlops/core
  • Total system peak performance: 158 TeraFlops/s
  • Top500: Rank 27 in 06/09
  • 20 TeraByte memory
  • 3 PetaByte GPFS file system (additional 3 PetaByte in 2011)
  • 8 x Qlogic 288 port 4 x DDR Infiniband switches
    • bandwidth: 16 GigaByte/s node to node (bidirectional)
    • 25.4 km Infiniband cables
  • http://www.dkrz.de/dkrz/about/hardware/HPC_Cluster

But Blizzard is connected to the Data archive:
  • HPSS: High Performance Storage System
  • 6 Sun StorageTek SL8500 tape libraries
    • 8 robots per library
    • 10 000 media slots per library
  • 73 tape drives
  • total capacity: 60 PetaByte.
  • 60.000 media (T10000 A/B, LTO4, 9940B and 9840C )
  • 500 TeraByte disk cache
  • projected fill rate: 10 PetaByte/year
  • bidirectional bandwidth: 3 GigaByte/s (sustained), 5 GigaByte/s (peak)
  • http://www.dkrz.de/dkrz/about/hardware/DataArchive
and the visualization machine:
  • 8 Vis Nodes (sva91 ... sva98)
    • HP XW 9400 Workstation
    • 2 Dual Core Opteron64 2.6 GHz
    • 32 GB RAM
    • 2 NVidia FX 5500 Graphics with 1 GB RAM
    • InfiniBand HCA
  • 1 SMP Vis Node (sva90)
    • HP ProLiant DL585 Server
    • 4 Dual Core Opteron64 2.4 GHz
    • 128 GB RAM
    • 2 NVidia FX 5500 Graphics with 1 GB RAM (via NVidia QuadroPlex)
    • InfiniBand HCA
  • 1 System Management Node (sva99)
    • HP ProLiant DL 385 Server
    • 2 Dual Core Opteron 2.2 GHz
    • 4 GB RAM
    • InfiniBand HCA
  • HP SFS (Scalable File Share) 72 TByte brutto / 48 TByte netto
    • 2 Meta Data Server (MDS)
    • 6 Object Storage Targets (OST)
    • Infiniband HCAs
  • http://www.dkrz.de/dkrz/about/hardware/vis_cluster

Comments

Post a Comment

Popular posts from this blog

Code evolution

I like computer generated graphics, but if can be generated by real data and can improve some insight I say: WOW look this! Last day I saw Michael Ogawa's works about how some codes had an evolution, I think he is using the data captured in the CVS, parsed and input to create a really amazing animations. He worked with Eclipse, Python, PostgreSQL and Apache, which are good examples of open source projects with many releases and people coding. Eclipse required HD because the enormous data. The videos: Eclipse HD Python PostgreSQL Apache From: SlashDot
Read more

Advice for potential graduate students

I found this text,copy and paste because it's an advice and must be in every biological laboratory. ADVICE FOR POTENTIAL GRADUATE STUDENTS We currently have room in the lab for more graduate students. But before you apply to this lab or any other, there are a few things to keep in mind. First, be realistic about graduate school. Graduate school in biology is not a sure path to success. Many students assume that they will eventually get a job just like their advisor’s. However, the average professor at a research university has three students at a time for about 5 years each. So, over a career of 30 years, this professor has about 18 students. Since the total number of positions has been pretty constant, these 18 people are competing for one spot. So go to grad school assuming that you might not end up at a research university, but instead a teaching college, or a government or industry job. All of these are great jobs, but it’s important to think of all this before you go to school...
Read more