UKAN

UKAN: Uncommitted KAOS Accessible Nodes

KAOS (Compilers, Hardware Architectures, and Operating Systems) is a systems research group at the University of Kentucky, externally known as Aggregate.Org. We operate multiple cluster supercomputers and are happy to freely share our computing resources with worthy users.

I Need Compute Cycles... Can I Use Your Machines?

If you just want to do production runs of a program written by somebody else, for which you have no involvement with the source code, the answer is almost certainly no. The answer is probably yes if:

  1. You have been or will be directly involved in creating or improving the source code of the application you wish to run.
  2. There is a compelling reason why University of Kentucky resources, and the volunteered efforts of members of our research group, should be used to support your project. Valid reasons are:
  3. Although we can treat resources as a sharable pool of nodes, we prefer to allocate an entire cluster to a single parallel program at a time.
  4. If our facilities are not sufficient, but easily can be upgraded, we may be willing to do so if you provide the parts. For example, we have little disk space, but adding disks costs only about $125 per 2TB drive. Covering costs associated with replacing failed parts, etc., also can make more nodes available for your work.

Given that volunteers maintain our equipment and we have little to no maintenance budget, we cannot make any promises about QoS (quality of service). In fact, we reserve the right to kick stuff off any machine at any time without prior notice. If you cannot live with this "best effort" level of service, UKAN is not for you.

What Do You Have And How Do I Use It?

Although the group has many clusters, not all are available to UKAN users at any given point in time either due to maintenance/configuration issues or because some machines are busy with our group's research. A reasonably up to date list of equipment immediately available to UKAN users is kept here.

Just to put this in perspective, as of December 9, 2012, UK deployed a $2.6M cluster supercomputer with a theoretical peak of around 140 TFLOPS using "nearly 5,000 central processing units and 48 high-performance graphics processing units." In contrast, the cluster supercomputers and other systems within our KAOS group at that time total a theoretical peak around 25 TFLOPS using about 400 conventional processor cores and 80 GPUs (and a cluster of FPGAs that don't contribute to the FLOPS count).

We have created a reference card (PDF) overviewing our equipment, using OpenCL with KOAP, and using MPI.

How Do I Apply?

Send your application as an email to UKAN at the University of Kentucky's engineering school, engr.uky.edu. Your email should include:

Any information submitted will be used only within our research group for purposes directly related to the UKAN facilities.

This Is Cool. Can I Help?

Of course! If you are a student at the University of Kentucky, you can particupate in any cluster "build parties," or perhaps get more involved by attending our group meetings and joining the KAOS list serve.

"Give me your tired, your huddled masses yearning to breathe free. The wretched refuse of your teeming shore." For us, that's about computer equipment. We are experts at making good use of old equpiment that was about to be trashed.

We also have a long history of getting new machines donated to us by manufacturers just before or shortly after public release. Of course, we love that too, and are very happy to serve as "alpha testers" and even to help create and refine the system hardware/software.

To help us with any of this, contact Professor Hank Dietz.


This page is: http://aggregate.org/UKAN/

The Aggregate. The only thing set in stone is our name.