A Beowulf-class supercomputer (or Beowulf cluster, as they are commonly known) is a high-performance distributed multiprocessor built from commodity off-the-shelf computers. These computers are connected together with a commodity network, so that they can communicate with one another. Each computer runs a free, open source operating system (often Linux), and typically provides a communications library like the Message Passing Interface or the Parallel Virtual Machine to make it "easy" for programs to run and communicate across the network.
High performance is achieved by writing the programs in such a way as to have them divide a problem into pieces and solve those pieces on different computers, in parallel.
A teraflop-scale computer is one that can perform at least one trillion double-precision floating point operations per second.
To illustrate, here are sixty integer problems. If you could solve them in one minute (by hand), you would be able to do one (integer) operation per second.
Now, here are sixty single-precision floating point operations.
If you could solve them in one minute (by hand), you would be able to do one (single-precision) floating point operation per second.
Last, here are sixty double-precision floating point operations. If you could solve them in one minute (by hand), you would be able to do one (double-precision) floating point operation per second.
A teraflops-scale computer can do at least 1012 of those operations each second!
3. How fast is Dahl?
Dahl's theoretical maximum speed (Rpeak) is about 3.8 teraflops.
We are still measuring Dahl's actual speed (Rmax), but we expect it to be between 2 and 3 teraflops.
A typical dual-core laptop may reach 1 Gflop (1 x 109 flops) of measured performance. So Dahl will be about 2000 times faster than that.
Desktop systems are faster than laptops, but not by very much -- maybe twice as fast -- so Dahl will be maybe 1000 times as fast as a typical desktop.
Please see our hardware page for a complete description of Dahl's hardware. Dahl's hardware cost roughly $165,000 (plus about $20,000 for a 65 kva UPS system, half of which Calvin paid for). The remainder of the grant so far has gone for software packages needed by Calvin researchers.
Dahl was paid for by the National Science Foundation (NSF). We gratefully acknowledge their support of this project!
Calvin College also contributed infrastructure, renovation, and release-time costs for the project.
Dahl is primarily used for faculty research projects involving computational modeling. Among the faculty using it are:
Dahl is also used for teaching undergraduate students how to program a supercomputer, in our High Performance Computing (CS 374) course.
2. What does teraflop-scale mean?
4. How does that compare to my computer?
5. What kind of hardware does Dahl use?
6. How much did Dahl cost to build?
7. Who paid for Dahl?
4. For what do you use Dahl?
| Calvin > CS > Dahl > FAQ |
|