harnesses Internet handshakes
Technology Research News
Whenever you click on a link on a Web site,
your computer sends a message to the site's Web server and the server
responds. Billions of such network handshakes take place on the Internet
every day. Although individually these handshakes are insignificant, large
numbers of them can add up to an impressive amount of computer processing
A team of researchers at the University of Notre Dame has figured out
a way to use Web
server handshakes to compute small pieces of a mathematical problem by
disguising the pieces as ordinary Web browser messages.
The researchers' parasitic computing scheme uses the processing power
of unwitting Web servers by exploiting one of the most basic operations
carried out by all computers connected to the Internet, the Transmission
(TCP) checksum, said Vincent W. Freeh, an assistant professor of computer
science and engineering at the University of Notre Dame.
TCP breaks messages from one computer to another into small pieces, or
packets, sends them over the network and reassembles them on the receiving
end. The TCP checksum adds up the number of bits in the message and attaches
the result to the message. On the receiving end, the computer adds up
the number of bits received and compares it to the checksum number to
make sure the message arrived intact.
The researchers performed their experiment with a type of math problem
that can only be solved by examining each possible solution until the
right solution is found. They encoded each candidate solution as a Web
browser request for a web page so that the TCP checksum was actually checking
to see if the message contained the correct solution.
The Web servers that received requests treated the messages that contained
failed solutions to the math problem as corrupted messages and discarded
them. The Web servers treated messages that contained the correct solution
as a request for a Web page that did not exist and sent the standard 'page
not found' error message to the researchers' computer.
Although the parasitic computing scheme demonstrates a principle, it is
not a useful tool because the amount of computer resources used to implement
the scheme far exceeds the amount that would be needed to solve the problem
on the researchers' computer by itself, said Freeh.
"For a general communication protocol, I think the probability [of developing
an efficient version of the scheme] is very remote," he said. "By design,
the receiver doesn't have to do that much. However, I think people are
[already] exploiting specific Web sites."
Web servers that run interactive applications and process forms are good
candidates for this kind of scheme, said Freeh. "This is where lots of
host cycles can be gotten," he said.
The parasitic computing scheme raises the possibility that computers on
the Internet can be used in ways their owners are unaware of, which raises
ethical and legal issues about the use of publicly available computer
Though the Web server resources used in the Notre Dame implementation
were barely measurable, if the scheme were used aggressively it could
have a similar effect to denial-of-service attacks in which one or more
Web servers are flooded with messages and effectively shut down, said
Ian Foster, a computer science professor at the University of Chicago
and a senior scientist at Argonne National Laboratory.
Each tiny message in the parasitic computing scheme is by itself indistinguishable
from any other Web page request, said Freeh. "The way to tell is by seeing
many such messages and deducing what is happening," he said. The researchers
have configured an intrusion detection system to detect their parasitic
computing scheme and they are working on configuring the system to detect
variations of the scheme, he said.
"The instance of parasitic computing that [the researchers] demonstrate...
is totally inefficient, returning a minuscule amount of computation for
great effort," said Foster.
The question is whether there are more efficient versions of such a scheme,
he said. "Within the Internet infrastructure [it] seems very unlikely
to me, given its fundamental simplicity." It's more likely, though still
doubtful, that someone could develop an efficient scheme to exploit peer-to-peer
networks like Gnutella, he said. "I wouldn't discount it totally, especially
as these infrastructures evolve."
Even if computationally efficient versions of the scheme can be developed,
it remains to be seen if it can perform useful work, said Miron Livny,
a computer science professor at the University of Wisconsin.
"It's a creative idea [but] it's not clear to me how it will work if you
really care about the result," said Livny. The the problem is the scheme
counts on messages that do not generate a reply to indicate that the message
did not containing the correct solution, but failing the TCP checksum
is not the only reason a message might not be returned. "The biggest challenge
in distributed systems is to understand why somebody is not responding,
because there [are] many, many reasons why you didn't hear back," Livny
The researchers tested the reliability of their scheme by repeatedly sending
out the correct solution. They got the correct answer back at rates that
varied from about 99 out of 100 to about 16,999 out of 17,000 times ,
according to Freeh.
Freeh's research colleagues were Albert-Laszlo Barabsi, Hawoong Jeong
and Jay B. Brockman of Notre Dame. They published the research in the
August 30, 2001 issue of the journal Nature. The research was funded by
the National Science Foundation (NSF).
TRN Categories: Internet; Distributed Computing
Story Type: News
Related Elements: Technical paper, "Parasitic Computing,"
Nature, August 30, 2001
Internet stays small world
Tools automate computer
Nanotube kinks control
Hydrogen chip to fuel
Research News Roundup
Research Watch blog
View from the High Ground Q&A
How It Works
News | Blog
Buy an ad link