|
发表于 2010-3-26 06:10:31
|
显示全部楼层
原文 文本, 有兴趣的就继续完成吧
Volunteer Computing Needs You
Distributed Computing Projects Drive New Ideas
Time to man up those motherboards, computer enthusiasts, and give some of that unused desktop power to those who really can put it to goods use.
Academic researchers are looking for a few good CPU cycles (and now even a few good human eyeballs).
Although fascinating distributed computing projects, such as: SETI@Home scour the skies for signs of E.T. phoning us, and GIMPS (Great Internet Mersenne Prime Search;www.mersenne.org) continues to find new prime numbers, an even wider group of researchers are starting to tap the voluntary computing grid to find cures for cancer and other worldwide humanitarian causes (World community Grid),look for new drugs (Docking@Home),and even track earthquakes (Quake-Catcher Network).
Distributed computing, which uses the spare CPU cycles on ordinary desktops to participate in massive computing projects, has become an important tool for researcher of all kinds.
---------
The most famous volunteer computing projects are still expanding.
The SETI@Home project, which looks for radio signals from space to indicate intelligent life, just launched a spin-off project called Astropulse, which also searches for pulsars and black holes.
Folding@Home, which studies protein folding mechanisms to better understand disease, reported in early 2009 that it now had combined enough desktop CPUs,PlayStation 3 Cell processors,and Nvidia and AMD GPUs to create 5 petaflops of computing power.
In fact, GPUs have become major contributors to some projects. Folding@Home says that Nvidia chips alone are responsible for 2 petaflops.
---------
It's that kind of cheap, massive computing power that is bringing distributed computing to a new level.
The technology is helping academics imagine research endeavors that would have been improbable a few years ago.
For instance, the Einstein@Home project is looking for pulsars (or spinning neutron stars) in space by processing date from gravitational wave detectors as well as data from one of the largest radio telescopes on earth, the Arecibo Observatory in Puerto Rico. The big telescope dish scans the cosmos for pulsars in five-minute increments.
---------
"Our raw data sets generate about 1,000 individual data sets, and each of those processes requires about 12 hours of computer processing," says Jim Cordes, professor of Astronomy at Cornell and a developer of the project.
"Five minutes of data takes about 10,000 hours of computing time to process. No typical university computer center can analyze that data to search for these kinds of objects."
---------
But it can be done via the 1 million host computers now participating in the project via the BOINC (Berkeley Open Infrastructure for Network Computing:boinc.berkeley.edu) platform, which was a by-product of the SETI@Home project.
Aside from Folding@Home, which has its own distributed computing platform, most academic projects now use the BOINC system to let your desktop CPU and GPU share cycles with an ongoing project.
--------
For Cordes, BOINC lets him send a small batch of observatory data to a single computer."[The receiving computer] churns away at [the data] for hours or a day and sends it back"
And there is no dearth of jobs to send.
Einstein@Home cruuently has 200TB of raw observatory data and plans to grow that to 1 petabyte.
The project has already confirmed and discovered pulsars.
-------
"Volunteer computing provides me with cheap access to a lot of computers," says Michela Taufer.
"Ican compute as much as I need."
Taufer, an assistant professor at University of Delaware's Department of Computer and Information Sciences, directs the Docking@Home project, which is testing new drug concepts by simulating how small molecules "dock" with proteins.
This requires simulating hundreds of thousands of potential moudels.
For a junior faculry member,acquiring an institution's own computing time requires lengthy applications and only small windows of actual usage on the college supercomputer.
But by deploying BOINC, she has 13,000 volunteers for Docking@Home dedicating 26,000 computers to the task.
------
"It is very tough to get all of this computation done easily," she says.
The emergence of a free and user-friendly system such as BOINC hase not only opened up new levels of computing power but also helped to fuel innovation.
"Volunteer computing provides me with cheap access to a lot of computers, so I can compute as mush as I need."
*******************************
People Power
The distributed computing concept has expanded in recent year to engage more than just computer cycles.
Now, some researchers are using humans themselves for distributed taskes.
In 2001, NASA initiated a "Clickworkers" project that had volunteers visually scan hi-res images of Mars to map craters.
In late 2009,the space agency joined with Microsoft to launch the Be A martian! online project(beamartian.jpl.nasa.gov/welcome) that turns Mars mapping into a game.
The modelresembles the Galaxy Zoo(galaxyzoo.org) project in the UK, where more than 150,000 people help review telescope images of the cosmos and classify the galaxies by their shape and features.
-------
"People are computers, too," says Landon Curt Noll, cryptologist and security architect at Cisco Systmes and a member of the Electronic Frontier Foundation's Cooperative Computing Award team.
"That kind of hybird approach is exciting. Astronomy has a massive amount of data just waiting for someone to analyze. You don't need to build an artificial intelligence engine, but you can take advantage of people volunteering time and using back-end computing to make sence of the research."
One hybird project that is contributing to our understanding of very down-to-earth events is the Quake-Catcher Network(qcn.stanfor.edu),which uses BOINC coupled with motion sensors to record earthquake activity. (See sidebar.)
"We have over 1,000 sensors in any given week that are sending us data back," says Elizaberth Cochran, assistant professor at the University of California, Riverdale.
Using motion detectors both in and attached to PCs clustered in California and Europe, the network of volunteers has gathered data on seismic events.
--------
"We have a number of records from the recent magnitude 6.5 in northern California," she says.
A number of Quake-Catcher sensors at Stanford picked up another lower-level magnitude 4.
"The value comes in getting a high density of sensors in places like Los Angeles and California."
*************************************************
Distributing Science
As many distributed computing project leader have discovered over the years, however, engaging people power of any sort requires more community outreach and communications skills than many academics are prepared for.
These programs succeed not merely because projects such as quake catching or galaxy mapping are cool.
--------
"[Success] is combination of user interest, funding, and the ability to communicate with volunteers," says Rom Walton, computer programmer/analyst at the University of California, Berkely's Space Sciences Laboratory and one of three full-time programmers of the BOINC infrastructure.
"If a project is unable to communicate with its volunteers, they go away and it ends up dying."
--------
One of the biggest programes associated with BOINC in the last year has been the Intel-sponsored Progress Thru Processors (www.facebook.com/progressthruprocessors) initiative, which tries to popularize distributed computing projects among everyday users.
--------
"The primary communication mechanism is Facebook," says Walton.
"People can see how their friends are doing in terms of how much computing time they've contributed."
--------
Walton sees social media as fertile ground for distributed computing because it maps so well against the distributed computing idea.
The computer grid is learning to use the online social grid.
People tell others about the projects that interest them and share bragging rights over how much they contributed to various computing causes.
Intel's Facebook page has 125,000 fans.
In coming years, the BOINC software may work directly with Facebook APIs so that new projects coming online with automatically create their own Facebook page and link and recruit users on the social networks the volunteers have already embraced.
----------
"Our idea is to make BOINC more social network-friendly," says Walton.
But in becoming more community-friendly, distributed computing projects are also learning to spread the word about science.
Cochran's Quake-Catcher project is reaching beyond seismic enthusiasts into K-12 classrooms.
----------
"We can provide educational activities and teachers with information about earthquakes to teach in class," she says.
The Quake Catchers can come to class and give live demonstrations of the sensor at work.
In exchange they ask the class to maintain it for a year.
"Any earthquakes the record can be shown to the class.", she says.
----------
Taufer believes that adding functions for communicating with the user base and sharing data on social networks through the BOINC platform will help turn some of thest distributed computiing projects into great opportunities to retain volunteers by educating them about science.
-----------
"The volunteers are more actively learning and becoming scientists," she says.
------------
The technology of distributed computing continues to make advances.
Walton says that BOINC now supports both Nividia and AMD GPUs, but it's up to each project to make greater use of this enormous untapped power.
Noll says that highly complex projects, such as calculating a prime number with 100 million digits, will require new algorithms that let many PCs work more cooperatively on a single large calculation.
The Electronic Frontier Foundation recently awarded $100,000 to a GIMPS member machine that discovered the first prime number with more than 10 million digits.
The next level of award, $150,000 for a 100 million-digit prime, was designed to force computer scientists to advance the art of distributed computing.
----------
"It is going to require that we either wait for machines to get fast enough or for people to get smart enough about how to make computers cooperate," says Noll.
----------
But the biggest advance and benefit for science in these volunteer projects could be in giving scientists a broader audience with whom to speak.
----------
"It has so many payoffs," says Cordes.
"It makes the case for our projects and for people being involved.
They are a great vehicle for education.
They attract people to Web sites to read and digest information.
It is just a win-win."
---------
In the end, researchers hope that what really is being distributed in voluntary computing projects is not just processing power but a deeper and broader understanding of science itself.
--------
by Steve Smith
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
Caught in the web
Corralling The Quake Catchers
The volunteer distributed computing model has inspired many researchers to consider new ways of utilizing the millions of connected PCs in the world - and not just for their sheer computing power.
Seismologist Elizabeth Corchran of the University of California,Riverside and her colleagues at other institutions developed the Quake-Catcher Network(qcn.stanford.edu).
One way this system detects earthquake activity is use the accelerometers that are already in many laptops to protect hard drive shock.
For desktop PCs, an inexpensive sensor sits on the floor and is connected to the PC via USB port.
The BOINC distributed computing platform then processes and sends the data to Stanford, UC Riverside's partner in this project.
Corchran explains how the technology works and with this is the kind fo project than can only be done affordably with dedicated volunteers.
-----------------------
CPU: You use BOINC differently from most other distributed computing projects.
Cochran: Most of the BOINC-related work is to read the acceleration data of of the sensors.
There is minimal computation.
It is lookinig at the incoming signal and seeing whether it is significant compared to the previous minute.
That is how we end up getting a trigger so we only send data back when there is significant movement.
The sensors using about 1% to 2% of the CPU.
----------------
CPU: How can you tell the difference between a quake and someone shaking their computer?
Cochran: We have a triggering algorithm that looks at the incoming signal and compares them to frequency bands typical of an earthquake.
But we really can't tell the difference between someone shaking their computer and someone in an earthquake.
The way we get around this is using a lot of sensors in a region.
---------------
CPU: So, in this case, a lot of the actual computing happens at your servers?
Cochran: All of these very small amounts of data go back to our server.
We send back the time, the amplitude, the sensor information, location,and type of sensor.
And we take all of that information in at our server and then quickly determine if the triggers are correlated in time and space.
---------------
CPU: What is this research telling you than traditional seismic observations can't?
Corchran: The distributed computing side makes it really easy to have anyone set up these sensors.
A lot of the problem in seismology is you have very complicated equipment that takes experts to set up.
This allows us to actually use reqular people to set us scismic stations for us and greatly expand the number of stations.
The BOINC software has made it really easy to set up a project.
In a few months, we were able to get some sort of platform running and get seismic data cominig back. |
|