About Me: Ankit Jain (ankit@berkeley.edu)

I am a third year undergraduate student in the EECS department focusing on Computer Systems.  My interests include High Performance Computing, Computer Security and Parallel Computing.  I have been researching under Professor Kathy Yelick and Professor James Demmel in the BeBop group.  I have been working on tuning Sparse Matrix Vector Multiplications on Scalar as well as Vector Machines.  

SETI@Home

Problem Description:

The Search for Extraterrestrial Intelligence was an initiative that started in 1997 and ended in 2005 in an effort to look for signs of life outside our planetary system by scanning the skies for any form of signal.  Due to the fact that the data being collected was being collected at a very fast pace and needed intense computation, scientists had to find a way to move the bottleneck from processing the data to collecting the data itself.  In 1999, a group at the Space Sciences Lab at UC Berkeley successfully launched SETI@Home which depended on volunteers on the internet to donate their processor power and time to process data that had already been collected and data that was being collected continuously.  Additionally, the development and success of this system sparked many other @Home projects such as Folding@Home and Einstein@Home which enable scientists to finish time and processor intensive computations in a reasonable time because of the distributed nature of the solution.[1]

Reason for Success:

The reason this model of distributed computing has been so succesful is due to the fact that each client can request a “work unit” that it will process independently of all other “work units” available.  During this process, the client and server do not need to interact much and thus the communication bandwidth bottleneck does not exist.  Given that users have varying internet connections and varying processor speeds such an effort would not have been possible for a problem which required intense client-server communication. 

Systems Used:

            The design of the SETI@Home client-server distributed computing model allowed the client to run on operating systems ranging from laptops to CRAYs’ using Parallel Virtual Machines technology that allows a heterogeneous collection of computers connected over a network, such as the internet, to behave like a single parallel computer. [2] Thus a user wanting to donate his/her idle CPU cycles towards computations of SETI@Home can run client software that enables the user to request “work units” and then return the result to SETI@Home servers. 

Scalability:

            A problem that collects 6.912/125 Terabytes of data a day needs a lot of computing power in order to process the data. This was achieved by having 412 thousand machines were working fulltime on the problem (CPU hours/hour) at any given time and returning results at 1 million WUs/day (thirty-four trillion calculations per second, roughly equivalent to a 249-teraHz super-computer). [3]

The scale at which this experiment is being conducted suggested a very simple error checking mechanism in which clients verify results from other clients. Due to the fact that there are so many computers and so many resources available, SETI@Home was succesful in implementing proper error checking algorithms based on the fact that if multiple clients return the same result, the result must be right.  This also prevents malicious software from being able to cause false positives in the search for ET.

Conclusions:

            Although there were no signs of Extraterrestrial Intelligence that were found during the 9 year study conducted, this project developed a system that is now being used to solve problems in Biomedical Research, Protein-Related Diseases, Climate Prediction and Emission of Gravitational Signals by Pulsars among many others.  This project also illustrated the power of distributed scientific computing to the now connected world and has foreshadowed possible ways for individuals to sell the computational power of their personal computers for use in such scientific experiments of the future.


 

References:

[1] SETI@Home: http://setiathome.ssl.berkeley.edu/

[2] Parallel Virtual Machines: http://www.csm.ornl.gov/pvm/pvm_home.html

[3] SETI@Home Statistics: http://www.roving-mouse.com/setiathome/