Architecture

POD is a high-performance, scalable, on-demand HPC environment with high-density compute nodes and directly attached, high-speed storage. POD users have their own personal login node which provides a persistent and secure compute environment. Jobs launched from the user's login node are executed directly on the compute node's physical cores and have access to the Direct Attached Storage. GigE, 10Gig and high-performance InfiniBand networks are available for compute node interconnects. Jobs run over a localized network topology to maximize inter-process communications and bandwidth, and to minimize latency.

Hardware

Compute Nodes

  • Dual-four core Xeon, dual-six core Xeon, or quad-twelve core AMD processors
  • Processor speeds ranging from 2.2GHz to 2.9GHz
  • 24GB to 128GB RAM per server
  • Up to 1TB scratch local storage per node

POD Can Be Rapidly Configured or Scaled

  • Dynamically provides the capacity you need
  • Contact us for specialized compute environments that can be configured

GPU Computing is Available via NVIDIA’s Tesla GPUs

  • Two NVIDIA Tesla C2075 Computing Processors per server
  • Each Tesla C2075 provides 448 cores and 6GB of GDDR5 memory with 144GB/s memory bandwidth
  • Server additionally provides Dual-four core Xeon Nehalem processors and 36GB RAM

Network

Internet Connectivity

  • POD has redundant high-speed Internet connections to allow you to easily move data into and out of the HPC environment. Depending on the customer configuration, connectivity can be anywhere from 50Mbps to 1Gbps. We work with customers to optimize data transfers across the WAN.

GigE/10Gig Ethernet

  • The standard POD HPC environment has all compute nodes connected via low-latency 10Gig or QDR IB networks. The network topology is localized to ensure maximum bandwidth and minimum latency between nodes offering good IPC performance. The storage systems are each made available via 10GbE to the local compute cluster.
  • Customized POD HPC solutions can offer full bisectional bandwidth on QDR Infiniband or 10Gig networks for large clusters. POD leverages Penguin Computing HPC expertise in deploying the best-performing networks to its customers.

Infiniband

  • Our Infiniband network offers the highest-performing, lowest-latency interconnect to meet the requirements of POD customers using the most demanding HPC software.

Storage

High Speed NFS

  • Our standard storage offering for POD HPC customers is through the PODStore systems. They offer high performance NFS access through 10Gig attached storage systems.

Parallel Filesystems

  • For more demanding and specialized workloads, POD HPC offers parallel filesystem options attached via multiple 10GbE links and Infiniband. Lustre and Panasas high-performance storage systems can be provided.

Hosted Storage Servers

  • Some customers require dedicated storage servers to isolate data and to facilitate encryption/decryption of high volumes of data being shipped rather than transferred over the network. These dedicated storage servers are made available to the compute cluster at wire speed to ensure efficient utilization of resources for the workloads. Because they are large-capacity, dedicated storage servers, they offer high security coupled with very low cost and ease of data movement.