60 likes | 240 Views
Virginia Tech NDSSL-HPC Proposed System Diagrams 06-11-2013. Diagram Page Architecture Diagram 2 Equipment Rack Placement 3 Components 4,5 Management Network 6. Racks 1-4. Rack 5. 10 GbE Network. 1GbE Management
E N D
Virginia Tech NDSSL-HPC Proposed System Diagrams 06-11-2013 DiagramPage Architecture Diagram 2 Equipment Rack Placement 3 Components 4,5 Management Network 6
Racks 1-4 Rack 5 10 GbE Network 1GbE Management Net Switches (stacked) Racks 1-4, each with: 18 x C2112 2U Quad Node Chassis 80 nodes each with: 2 x E5-2670 CPUs 64GB Memory (4GB/core) 1 x 1TB HDD FDR IB 2 x 48-port 10GbE Switches FDR IB Fabric 432 leaf ports expandable to 648 ports Racks 6 1GbE Mgmt switch 4 x C1104-RP5 1U Xeon PhiNodes 2 x E5-2670 CPUs 64 GB Memory (4GB/core) 1 x 1TB HDD 2 x C2108-RP2 Admin nodes 2 x Intel Xeon E5 2670 2.6 GHz 128GB Mem 2x1TB HDD RAID-1 1 x PCIE-CONNX3-1P (FDR IB) 2 x 10GbE ports (SR) Console/KVM 2 x C2108 VM servers, 2 x Intel Xeon E5 2670 2.6 GHz 128GB Mem 3x1TB HDD RAID-1 1 x PCIE-CONNX3-1P (FDR IB) 2 x 10GbE ports (SR) 2 x C2108-RP2 Login nodes 2 x Intel Xeon E5 2670 2.6 GHz 128GB Mem 2x1TB HDD RAID-1 1 x PCIE-CONNX3-1P (FDR IB) 2 x 10GbE ports (SR) 1 xC2108 archive node 2 x Intel Xeon E5 2670 2.6 GHz 128GB Mem 2x1TB HDD RAID-1 1 x PCIE-CONNX3-1P (FDR IB) 2 x 10GbE ports (SR) 2 x C2108-RP2 Gateway nodes 2 x Intel Xeon E5 2670 2.6 GHz 128GB Mem 2x1TB HDD RAID-1 1 x PCIE-CONNX3-1P (FDR IB) 2 x 10GbE ports (SR) 4 x UV20 fat nodes 4 x E5-4650 2.6GHz 8 core 1.5TB memory, 1 x 1TB HDD, FDR Racks 9 1 x 1GbE Mgmt switch 4 x UV20 Oracle Servers each with 10 x 3TB HDD NFS Server 1 MIS Server with 72 3TB HDD 1 MIS JBOD Expansion with 72 3TB HDDs Racks 7,8 Cable Legend Racks 7,8 each with: 1 x 1GbE Mgmt switch 20 x C2110-RP5 2U Xeon Phi Nodes 2 x E5-2670 CPUs 64GB Memory (4GB/core) 1 x 500GB HDD 1 x 5110P Xeon Phi accelerators FDR IB 10GbE Network 1GbE Management Network FDR Infiniband Fabric NDSSL HPC System Architecture
C2112-4RP4 2U Quad Node Chassis • 4 x compute nodes in 2U, each with: • - 2 x E5 CPU sockets • 16 x 1600/1333/1066/800 DIMM Slots • - 3 x 3.5” HDD per server • 1 x PCIe 3.0 x16 supportint low profile slot • 1 x FDR Infiniband port on board • - 2 x 1GbE NICs • - 2 x 1200w redundant power supplies C1104G-RP5 Accelerator node 1U 2 x E5 CPU sockets 8 x 1600/1333/1066/800 DIMM Slots 4 x 2.5” HDD 3 x PCIe 3.0 x16 supporting 3 GPGPU or Xeon Phi cards 1 x PCIe 3.0 x8 slot 2 x 1GbE NICs 2 x 1800w Platinum level redundant power supplies • C2108 • - 2U dual-socket E5-2600 series server • - HPC management, head node, I/O nodes • - Up to 8 x 3.5” hot-swap drives • - Up to five expansion PCIe slots • - 1+1 Redundant AC Power Supplies • UV20 2U Quad Socket Server • - 4 socket E5-4600 series server • - 48 DIMM slots (1.5TB mem with 32GB DIMMs) • - Up to 8 x 2.5” HDDs • 8 PCIe slots • Redundant power NDSSL HPC Components
SGI IS2112 2U 12-drive JBOD • 12 3.5-inch HDD • SAS Interfaces SGI Modular InfiniteStorage Server • Storage Server Options • Integrated into 4U chassis • 1 or 2 motherboards (Sandy bridge Platform) • 2 socket LGA ES-2600 (E5-2600 Series) • Each chassis supports SATA/SAS/SSD drive types: • Up to 72 x 3.5” drives • Up to 144 x 2.5” drives • Dual GbE onboard. Optional 2 port 10GbE, 2 port GbE, or 4 port 8Gb FC PCIe cards • JBOD Option • Up to 81 SATA/SAS/SSD drives (162 w/2.5” SFF) • Up to 8 SAS connections NDSSL HPC Components
Rack 1 Rack 2 Rack 3 Rack 4 Rack 5 Rack 6 Rack 7 Rack 8 Rack 9 Racks 1-4 each with: 19 x C2112 Quad-node compute chassis (total: 304 nodes) Rack 5: IB switches, 10GbE switches Rack 6: 4 x C2112 Quad compute chassis (16 nodes) 2 x admin nodes, 2 x VM servers, 2 x login nodes, 1 x archive node 4 x UV20 fat nodes Rack 7-8: each with: 40 C1104-RP5 1U nodes with Xeon Phi ( total: 80 Xeon Phi nodes) Rack 9: 4 x UV20 Oracle servers with local storage NAS Storage: 1 x MIS server with 72 x 3TB HDD + 1 x MIS JBOD expansion with 72 x 3TB HDDs Total: 144 x 3TB HDDs = 14 x 8+2 LUNs + 4 hot spares = 319.2TB usable capacity NDSSL HPC Equipment Rack Placement
1G sw1 1G sw1 1G sw1 1G sw1 1G sw2 1G sw2 1G sw2 1G sw2 Bond0 Active-Backup Admin Nodes 10G Nodes x 8 10G Nodes x 8 10G Sw1 10G Sw2 1G sw1 1G sw1 1G sw1 1G sw1 Rack 1 Rack 2 Rack 3 Rack 4 Rack 6 Rack 7 Rack 8 Rack 9 Rack 5 not shown 10G switches are in rack 5 NDSSL HPC Management Network