The Borg System

February 16, 2004

Borg was decommissioned today, Monday February 16, 2004 and will not be put back into production. Now, high performance computing (HPC) at ACCC will be done on a cluster of machines called argo.

Borg was a shared memory multiprocessor system: there is a single memory system with 4GB and the fourteen processors share it among the user and system processes. The operating system was the HP variant of UNIX.

Argo, is very different; comprised of 36 separate machines, it is a distributive shared memory multicomputer system. A program or software package runs as one or more processes on one or more machines. The processes running on multiple machines communicate with one another via a private network. Each machine is a fully functional computer with its own processor, memory, disk, and network connection(s). The OS is RedHat Linux.

Argo has a single head machine and thirty-two compute machines. (In cluster literature, the words "machine" and "node" are interchangeable. The head machine is also referred to as the "front-end" or the "master;" the compute nodes, "workers" or "slaves.") A client logins into the head node and has access to editors, compilers, and other tools (software, libraries, profilers, etc.) for serial and parallel program creation. Programs, user written or commercial, are run on the compute nodes under the control of a batch system called PBS. From the head node, clients can monitor the progress of a job and access results; they are, however, not permitted to login to compute nodes.

Clusters come in all shapes and sizes; it may be composed of just two machines (a head node and a compute node) or thousands of machines. The machines may be identical in hardware or dissimilar. A single processor or multiple processors (SMP) per machine is allowed. The type an quantity of memory among the machines may be the same or different. Multiple networks that connect some or all of the machines are acceptable. A hard disk on compute nodes is optional and network booting quite common. The National Center for Supercomputing Applications (NCSA) at UIUC has just made its Tungsten cluster available. It's VERY big; it has 1,280 machines with dual 3.0GHz Xeon processors. While argo is considerably smaller, many of the principles underlying both systems are the same.

Machines may be added to the argo system and older, obsolete ones, replaced. Machines may be configured with different hardware resources for different applications. While some machines may be brought down for repair or replacement, other machines will remain up, unlike borg where the entire system had to be brought down for maintenance. These are just a few of the benefits of argo which is considerably more flexible and cost effective than the borg system.

For more information about argo, see its home page.

If you have any questions that are not addressed on the argo web page, send them to systems@uic.edu.

DO YOU HAVE ANY QUESTIONS?

Send your questions, problems, and requests to:

systems@uic.edu