SETI@home is one of the highest-profile proof-of-concept projects for distributed computing. However its Berkeley-based administrators are, participants claim, ignoring allegations that cheating in the competition to contribute the most computing power to the project is rife.
With the competition's close just two months away, observers believe that long-time leader ARS Technica-sponsored Team Lamb Chop (ATLC) will lose its lead to relative newcomer to the leader-board, SETI@Netherlands, in 10 days. Competition veterans and even SETI@Netherlands own manager think the team's late burst through the ranks is a little too good to be true.
Opened in June 1999, the competition is a battle fought with CPUs and bandwidth. Each day SETI@home's headquarters in Berkeley serves millions of bytes of digitised space noise to the project's volunteers for decoding in chunks known as work units or WUs. Winning the competition is simple: return the most WUs.
ATLC has contributed a staggering amount of processing power to SETI@home, having returned over eight million WUs. By May, ATLC had returned around six million WUs giving it a three million WU lead over SETI@Netherlands. Since July SETI@Netherlands production has accelerated sharply, closing the gap to under a million work units.
Max Nealon is an IT professional who has followed the SETI@home project since its early beta days and worked with Berkeley to uncover instances of cheating. He says the probability that all of SETI Netherland's statistics are legitimate is highly improbable.
Nealon estimates that a 1GHz PC dedicated to SETI@home processing would take six hours to complete a single work unit. He said some members of Team Netherlands are returning 5000 WUs per day.
Nealon says that would mean team members producing this much work must have 1250GHz of processing power at their disposal dedicated purely to the project. In human terms, that's around 1,250 1GHz computers doing nothing but running the SETI@home screensaver.
"[SETI Netherlands] number one producer has 618,000 units, that's just one person, it's just ludicrous," said Nealon.
Nealon knows the kind of effort required for one individual to break into five figures in the competition let alone six.
"I've got 43,000 spread out over two accounts," said Nealon. "In order to do that I've brought laboratories of computers...at one point I had 35 computers in my garage all purely dedicated to SETI
"It's incredibly difficult".
However, according to Nealon, SETI@home's administrators don't want to know about it. Nealon has contacted SETI@home concerning the cheating several times but is yet to receive a response.
Nealon believes that SETI@home's resources have become so scarce that they're unwilling to deal with the problem without a very good reason.
"It's not important until such time as it comes to finding another sponsor," said Nealon.
The distributed computing infrastructure behind SETI@home will soon be replaced by Berkeley Open Infrastructure for Network Computing which will support the project's second phase, SETI@home II. However, according to Nealon, the cheating problem affecting SETI@home brings the amount of computing power and the accuracy of the results the project has attained so far into question.
Nealon said that there are several ways to cheat the SETI@home's statistics counters.
One common technique used by cheats is to distribute partially completed work units to other team members' SETI@home accounts. One account is used to process a work unit until it is 99 per cent complete. It is then distributed hundreds of other team members who process the remaining portion of the unit and return it. The WU is credited to their accounts vastly inflating the quantity of public processing that appear to be dedicated to the project.
A bulletin-board posting allegedly from SETI Netherlands' team manager states cheating throughout the team is rampant and estimates that 41 per cent of the team's work is illegitimate.
Nealon has also identified ways in which the cheating could pervert the accuracy of data received by the project. He said that it could undermine the mechanisms built in to the project to assure its accuracy.
"Every unit is processed a number of times until they get the same answer. If, for example, that number of times is three and they get three identical results from people who are sharing the same cheat, and that result is flawed, then potentially they are getting flawed data" said Nealon.
"When you start talking about three million results that are potentially flawed..."
It's also clear that those who have supported the project since its inception feel that they're being treated with contempt.
"Basically, three years of work to get to the top of the teams and eight million WU later, it looks as though the top team is going to be beaten by cheating," said Nealon.
"The people are only cheating because they can and there have been an awful lot of people who have spent an awful lot of time crunching numbers legitimately," he added later.
It is really unforunate that cheats may be the demise of the most popular DC project in the world. I know at TeamPhoenixRising.com we believe in crunching real work units. Our team has reached #24 in the ranks by hard work and honest endeavors.
As far as the stats go, it is impossible to compete with someone who subverts the scoring system.