Introduction

• Nature of the Computer

  The Vnode Cluster is a collection of twelve roughly identical commodity computers. Each node has it's own system image and a large main memory that is entirely separate from the rest of the cluster. This type of cluster is programmed with message passing to take advantage of parallelism.

  Each node is also equipped with a high-end Graphics Processing Unit (GPU). Individual GPUs can be programmed with a shader language, like NVIDIA C for Graphics, or a high-level language for general purpose GPU programming, like BrookGPU. As a cluster, The GPUs can be arranged and be programmed with Chromium.

• How do I use it?

  Computational jobs on the cluster are scheduled and run in a space-shared, exclusive access fashion. This means that when your job is running on a node, no other jobs are running on that node at the same time. In order to achieve this, we use two job queuing services, PBS and Condor. These services use information that you submit about your job such as the number of nodes and the amount of time that you request to schedule jobs in as fair a fashion as possible. The usual use pattern of the cluster is as follows:

  1. Submit a job
  2. Job scheduler schedules jobs on multiple nodes
  3. Job scheduler starts API on single node
  4. Parallel API starts process on allocation
  5. Your process runs and exits
  6. Job scheduler reports results

• What the cluster is not

  The Vnode Cluster is not a shared memory machine. This means that one needs to use explicit parallelism beyond multithreaded code in order to take advantage of the cluster. The cluster will not parallelize jobs for you.