LC3 Project – Low Cost Computer Cluster

LC3 (L C^3 = Low Cost Computer Cluster) was developed aimed at the using of high development processing through people or institutions that have limited financial resources.

One of the main reasons that encouraged the beginning of Beowulf computer’s clusters construction was the value. That was an impracticable value, even for big corporations, as the National Aeronautics and Space Administration (NASA) and other research centers around the world. This is from a long time and, since 1994, when NASA started the Beowulf project, many things have changed, but there are still some countries that have difficulty to get generic hardware, due to economic reasons or restricted import policies.

LC3 project was created to make a high performance computer cluster construction feasible, aiming at the minimum use of redundant component and the use of second-hand hardware, like Pentium’s first generation processors.

For this kind of solution, the GNU/Linux is perfect, since, besides providing all the technical requirements for the cluster’s working, it’s philosophy is similar to the LC3 project point of view, as this operational system aims the social and technological equality.

Description

Two computer networks compose the project structure. One is for the nodes exclusive use (Switch 100Mbps) and the other (HUB 100Mbps) is used for the client stations’ and the Internet gateway’s connections. The Front End server makes the connection between the two networks, as the picture below shows:

LC3 General Diagram

The node stations don’t have hard disk and work with the LTSP resources.

The Front End receives the client stations processing requests and then creates a process, which is divided for the node stations. As soon as the nodes conclude all the tasks, the results are collected by the Front End and delivered to the client station that ordered the task.

Working

The general working of the system needs, besides the mentioned hardware, many software where we can mention the GNU/Linux operational system, the exchange message library (PVM, MPI, etc.), the LSTP for node stations’ remote boot, and the applications, like software for mathematics, statistics, image render, simulations, decrypting and many others.

For HPC resources’ use by one client station, it needs to connect to the Front End and from the Front End, accomplish the processing requisition to the nodes. This is done through programming, by the use of C or Fortran languages, and the addition of message exchange libraries to the program’s code.

Results

On the performance tests, the LC3 cluster was really efficient. There were many tests, and the one that showed a superior performance was the image render by using the POV-Ray software. The PVMPOV is a POV-Ray’s software version that was written specifically for PVM clusters.

Picture: skyvase.pov

The tested image for the image render was the skyvase.pov, since it is a benchmarking pattern for the POV-Ray.

We can see, below, the performance chart, where the complete cluster got only 42 seconds to render all the image, with a better performance than a station equipped with a Pentium 4-1,5GHz processor, that takes, on average, 50 seconds to conclude the image render.

Equipment performance
Item Equipment
P100 Pentium 100MHz
P166 Pentium 166MHz
P200SMP(2) SMP Pentium 200MHz dual
LC3+2P100 Front End: SMP Pentium 200MHz dual;Nodes: 2 Pentium 100MHz
LC3+2P166 Front End: SMP Pentium 200MHz dual;Nodes: 2 Pentium 166MHz
LC3 Full Complete LC3 Cluster.Front End: SMP Pentium 200MHz dual;

Nodes: 2 Pentium 100MHz, 2 Pentium 166MHz

Performance Chart Reference Table

Conclusion

What makes the use of a cluster acceptable, instead of using a station with a Pentium 4 processor, as the cluster takes up more space, spends more power and needs a special way of programming?

The answer for this question is the cost of a cluster like this, that is much cheaper than a Pentium 4 station!

Nowadays, the price of a computer with a Pentium 4 processor is about US$ 520,00, while the implementation of a LC3 cluster costs about US$ 330,00.

In addition, it is known that electronic components are usually priced in US Dollars. In some countries, like Brazil and many others, the exchange rate for the American currency is always getting higher, due to economic instability. Therefore, a LC3 implementation is a very good way for reducing costs, since first generation Pentium equipments do not have a significant market value and can be bought by very low prices.

There are also the technological limitations, as this performance would not be able, with only one processor, three years ago, when nobody could believe in a processor with the Pentium 4 capability, as well as the processors like Pentium 100MHz were already obsolete.

Briefly, we can find that the greatest advantage of this high performance architecture use is the low costs, but are advantages above that. The use of clusters with free software can break technological restrictions, so that an underdeveloped country researcher is able to achieve the same processing capability as another researcher in a top research center like the Massachusetts Institute of Technology (MIT).

In this project, the use of GNU/Linux is essential, since the closest private software to it (UNIX like) would have an extremely high cost, due to the software copyright, and because these software would require higher level hardware. That is the reason why the GNU/Linux is becoming widely used in digital projects wide spreading.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: