Penguin Computing on Demand (or POD) is an HPC as a Service offering for high performance computing customers. POD offers end-users and service providers a high performance, on-demand and scalable HPC computing environment accessible from a workstation or as a virtual extension to an existing HPC cluster. POD is an excellent choice for researchers, scientists and engineers who require surge capacity for time-critical analyses or for small to medium-sized organizations that need high-performance computing capabilities without the capital outlay and effort associated with acquiring their own HPC cluster.
Overview
POD is a high-performance, scalable, on-demand HPC environment with high-density compute nodes and direct attached high-speed storage. POD users have a persistent and secure compute environment that executes jobs directly on the compute nodes’ physical cores. Both GigE and high performance Infiniband networks are available as the compute nodes' interconnect. Jobs run over a localized network topology to maximize inter-process communication (minimum latency and maximum bandwidth).
Accessing POD
POD users connect to a personal Login Node on POD by remote access (Internet or VPN) at an IP address provided by Penguin. This connection is verified against the IP address (or range of addresses) of the user for security. All user access is accomplished either through SSH using the login information provided by Penguin (standard Linux shell) or through a user-supplied Web-based user interface. The POD environment can host Web servers so a user can provide a customer-facing portal that is directly connected to their Login Node and to the HPC resources.
Using POD
Penguin's POD support team is always available to assist in configuring your computing environment on your Login Node. This environment will be persistent on POD for so long as you maintain a POD account. The Login Node will be configured with either RedHat or CentOS Linux, CentOS Linux modules and other tools, as well as a scheduler (either TORQUE Resource Manager or Sun Grid Engine).
Submitting Jobs
Once your computing environment is set up, you can submit jobs to be processed by POD. Jobs go into a queue and move through the queue as determined by the POD's job scheduler using a “Fair Share” policy. You can submit a query to POD to see the status of your jobs in the queue. Please note that if you are using SGE, then the name of each user's job may be visible to every other user of the POD service.
Storing Your Data on POD
Persistent storage is provided for storing your local data. POD provides both on-demand storage, where you pay only for what you use, and the option for longer-term, personal storage. The on-demand storage is attached to the compute cluster via several 10GigE network links to provide high I/O capabilities.
Each compute node has a 2TB scratch disk for local storage of data during compute. A Panasas parallel file storage system is also available for applications with many clients that need simultaneous access to a common data set.
Support
POD users receive technical support from Penguin’s HPC experts, including application set-up, creation of the HPC computing environment, ongoing support, data exchange services and application tuning.