Infrastructure

Core hours are the basic compute unit of POD. When a job is launched on POD using one of the job schedulers, one or more processes are started on POD's compute nodes depending on the parameters used to launch the job and the characteristics of the application. A Core Hour is defined as the use of one processor core by one of these processes for one hour. Note that it is sometimes useful to submit your jobs based on nodes (processors) rather than cores - for example if you know that you need the full 32GB of memory for your job. In this case, you will be billed for the entire processor (x4 cores) even if all the cores are not fully utilized.

Dedicated clusters can be used for continuous large workloads over a long period of time or when you have a need for a greater level of security. Dedicated clusters are reserved solely for your use - they sit idle when you are not using them. Dedicated cluster are generally provided on a monthly basis, although longer terms are available.

Data Storage

POD provides several options for your data storage needs. The options are tiered based on access speed, convenience and price. Regardless of the option you choose, POD storage is all directly attached to the compute node network so you are guaranteed low latency, high-speed access to your data.

• Local Scratch is available on each compute node. POD compute nodes have a 2TB local disk that you can use during compute as scratch space. Your scratch storage is included at no additional charge.
• POD Store is the basic bulk storage system of POD - a network attached NSF storage array connected to the compute network via a 10Gig network. POD Store is provided on a fee per average TB stored during the month, calculated on a daily basis. It is intended for medium duration storage requirements. A RAID10 storage device providing maximum reliability of your data, the POD Store system can be rebuilt if up to two disk drives fail at the same time, however it is not mirrored or backed up to disk/tape.
• POD Fast Access is a high-speed, parallel file system intended for fast access of data during compute runs. It is intended for applications that require multiple jobs to simultaneously access the same data file or for applications that need faster scratch space access than the local hard drive can provide. POD Fast Access is connected to the compute network via multiple 10Gig connections and is provided on a fee per average TB stored during the month, calculated on a daily basis. A RAID5 device, POD Fast Access can be rebuilt in the event of a disk failure, however if is not mirrored or backed up to disk/tape.

Login Nodes (including 3D Graphics Accelerated Nodes)

Each POD account requires a Login Node which provides a persistent, secure compute environment for the users of that account. The Login Node provides access to POD's physical compute resources including storage, master nodes and compute nodes. The Login Node's environment can be used for simple testing, access to usage and billing records, managing storage and compiling applications. Use of the Login Node is not included in the Core Hour fee calculations for POD - all charges associated with the Login Node are included in a single monthly fee.

Both Virtual and Physical Login Nodes are available. A Virtual Login Node provides the user with one core of a Xeon 5420 (2.4GHz) processor, 2 GB RAM and 900 GB of bandwidth per month. A Physical Login Node provides a quad-core Xeon 5420 processor, 2 GB RAM and 900 GB of bandwidth per month.

Also available are Physical Login Nodes that are equipped with a Quadro FX Graphics adapter from NVIDIA. Consider this node if remote visualization is important to your POD work flow. The POD support team can help you with configuring TurboVNC and VirtualGL in your environment. Network bandwidth charges do apply to your use of TurboVNC and VirtualGL.

Combination Storage/Login Nodes

A POD Combination Storage/Login Node is intended for users that plan on storing data on POD for many months, may be transferring data to/from POD on a regular basis, or may want additional data security and need an economical solution. A POD Combination Storage/Login Node is a physical server that provides a persistent user environment plus a number of empty drive bays that can be populated by the user as needed. For example, you could send four 2TB disk drives to your POD Combination Storage/Login Node (via overnight courier) with software RAID and encrypted data which will be installed in your server within hours after the disks arrive. You would then have instant access to 5TB of your encrypted data.

Dedicated Compute Nodes

Dedicated compute notes are available as an option and are useful if you know you will have continuous workload over a longer period of time or you have a need for a greater level of security (dedicated compute nodes are reserved solely for your use - they sit idle when you are not using them). The cost per core hour for a dedicated compute node is less than for cores available on demand. Dedicated compute nodes are generally provided on a monthly basis, although longer terms are available. Consider dedicated compute nodes if you know that you will need all of the memory available on a server for your application, regardless of the number of cores your jobs will require.

GPU Compute Nodes

GPU Compute Nodes are available on POD. Each GPU Compute Node has a dual Xeon X5540 4-core processor (2.53GHz), 24GB of memory (3GB per core), and 3 NVIDIA Tesla C1060 Computing Processors. GPU Compute Nodes are charged on a per server-hour basis rather than a per core hour basis.