A Linux-based "flash mob" supercomputer
From: | "Jessica Mullens" <jessica_mullens-AT-att.net> | |
To: | <jessica_mullens-AT-att.net> | |
Subject: | Linux Democratizes Supercomputing | |
Date: | Wed, 24 Mar 2004 14:15:52 -0800 |
San Francisco, March 24, 2004 On April 3, 2004 hundreds of computer enthusiasts will create FlashMob I, the first FlashMob Supercomputer in the world, and attempt to break into the Top500 Fastest Supercomputers on earth. Sponsored by University of San Francisco, FlashMob I will be created on the University of San Francisco (USF) campus, using USF student and faculty-designed open-source software. A FlashMob Supercomputer is created by connecting a virtually infinite number of computers via a high-speed LAN, to work together as a single supercomputer. A FlashMob computer, unlike an ordinary cluster, is temporary and organized ad hoc for the purpose of working on a single problem. It uses volunteers and ordinary laptop PCs, and is designed to allow anyone to create a supercomputer in a matter of hours. "This is a radical new idea in supercomputing, as well as an important scientific and social experiment, said John Witchel, graduate student and co-creator of FlashMob Computing. The goal of the FlashMob I project is to demonstrate the viability of widespread supercomputing. We hope to give ordinary citizens the power to explore and address problems that are most important to them whether its a high-school science class looking to participate in study of global warming, or a family impacted by breast cancer, or even a chess club looking to build an electronic grand-master. In short, we want to democratize supercomputing. "FlashMob Computing was effectively invented in the course of a classroom discussion, said Pat Miller, USF lecturer and computer scientist at the Center for Applied Scientific Computing at the Lawrence Livermore National Laboratory A student posited that could break into the Top 500, by simply inviting people to come together to create a supercomputer for a few hours. "Today, supercomputers are reserved for issues the government deems important, primarily defense research, said Greg Benson, USF Professor. We envision a supercomputing landscape where individuals can effectively vote with their laptops by volunteering to lend their computers to causes that are important to them regardless of governmental interest and support. Because FlashMob supercomputers are relatively easy to set up and its code is so portable, scientists can develop can write programs that are flashable, and put out a request for a FlashMob, breaking sciences long-time dependence traditional supercomputer centers. Our ability to harness the power and promise of supercomputing will directly impact our individual and global well-being and prosperity. FlashMob I FlashMob I will occur on April 3, 2004, at the University of San Francisco gym. Network setup will begin on April 2. Doors open at 8:00 AM, with arrival times staggered throughout the morning. Participants will be assigned a specific arrival time when they register. As participants arrive, they will be given a copy of FlashMob I Software which they will boot from their CD-ROM. Because the program runs entirely from the CD-ROM, hard drives will never be touched, safeguarding personal data security. Once the PC is connected to the network, it will self-configure and begin processing requests from a central server. With the addition of each volunteer computer, the power of the supercomputer grows. At 1:00 PM on April 3, USF students and faculty will run LINPACK, a well-known benchmarking software package. The last benchmark will be run at 5:00 PM that same evening. The best benchmark will be submitted for inclusion in the Top 500 Supercomputer list. Throughout the day, USF will host a series of lectures and round table discussions, with industry experts from Microsoft, Hewlett-Packard, NASA, NERSC and others, as well as technology exhibitions and an XBox tournament. FlashMob I is a modified Linux kernel containing original software that allows individual PCs to join a network and operate as a single supercomputer. Standard supercomputer libraries such as MPI have been specially tuned for the unusual nature of a FlashMob and original code has been written to facilitate bootstrapping PCs, real-time reporting, on the fly network and node diagnostics, and ad-hoc performance optimization. Both the modified Linux kernel and the accompanying software are burned onto a bootable CD-ROM and duplicated in volume. In the tradition of Open Source computing, raw data from the experiment will be made publicly available at www.flashmobcomputing.org, a website where people can share ideas, tune software, and improve implementation, to encourage further research in the area of FlashMob Supercomputing, and to provide a strong baseline of practical experience in preparation for FlashMob II and FlashMob III. Additionally, the site will start tracking FlashMob computers with a Top 500 Flashes list -- separate from the top500.org list to encourage the exploration of this exciting new area. ># # # About FlashMob I On April 3, 2004 hundreds of computer enthusiasts will create FlashMob I, the first FlashMob computer in the world. Sponsored by University of San Francisco, FlashMob I will be created on the USF campus, using USF student and faculty-designed open-source software. A FlashMob supercomputer is created by connecting virtually an infinite number of computers via a high-speed LAN, to work together as a single supercomputer. A FlashMob computer, is temporary and organized ad hoc for the purpose of working on a single problem. It uses volunteers and ordinary laptop PCs, and is designed to allow ordinary folks to create a supercomputer in a matter of hours. For more information, or to sign up to participate in this historic event please visit http://www.flashmobcomputing.org.
Posted Mar 26, 2004 7:17 UTC (Fri)
by blayne (guest, #19468)
[Link]
I think LINPACK tests the interprocessor connections as much as the processors, so it isn't as simple as summing the processing power of all the processor nodes. A LAN connection may not be up to the task. Real supercomputers certainly don't use ethernet for interprocessor communications. Even if we ignore interprocessor communications bottlenecks, there are many supercomputing tasks that require frequent exchange of data between nodes. When that's the case, the supercomputing speed approaches n times the speed of the SLOWEST processor, where n is the number of processor nodes. Adding a slow laptop could slow the entire supercomputer. To encourage more students to participate, the flashmob supercomputer sets a fairly low minimum hardware requirement. However, there are some supercomputer applications that process data independently and asynchronously. A good example is SETI@home. These applications can benefit from something like a flashmob supercomputer. If network connected distributed supercomputers were allowed on the top500.org list, SETI@home would be #1 by a large margin. It's going to be a lot more convenient to move the data instead of the processors. Several companies are looking at the success of SETI@home and spinoffs like Folding@home. They're working on commercial grid computing solutions that are similar to the flashmob supercomputer without the need to coordinate people to assemble with their laptops. Grid computing can use widely dispersed computers across the internet, or it can allow a large company to run supercomputing jobs on their in-house PCs when they aren't being used. But who can argue against any project that might improve the social skills of nerds? Good luck, Flashmob I!
Neat idea. Should be fun.A Linux-based "flash mob" supercomputer
References:
LINPACK: http://www.top500.org/lists/linpack.php
SETI@home: http://setiathome.ssl.berkeley.edu/
Folding@home: http://www.stanford.edu/group/pandegroup/folding/