The Leading Source for Global News and Information Covering the Ecosystem of High Productivity Computing
August 20, 2008
SAN DIEGO, Aug. 19 -- Computer scientists at the UC San Diego's Jacobs School of Engineering have proposed a new way to build datacenters that could save companies money and deliver more computing capability to end-users.
"Large companies are putting together server farms of tens of thousands of computers -- even approaching 100-thousand, and the big challenge is to interconnect all these computers so that they can talk to each other as quickly as possible, without incurring significant costs," said Amin Vahdat, a professor of Computer Science and Engineering (CSE) in UC San Diego's Jacobs School of Engineering. "We are proposing a new topology for Ethernet datacenter connectivity."
The innovation is outlined in a paper, titled "A Scalable, Commodity Data Center Network Architecture," presented today by Vahdat at the annual meeting of SIGCOMM, the Special Interest Group on Data Communications. SIGCOMM is the premier academic conference for researchers in the fields of communications and computer networks, and the event runs through Friday in Seattle, Washington.
Vahdat, who also directs UCSD's Center for Networked Systems (CNS), co-authored the paper with two CSE graduate students, Mohammad Al-Fares and Alexander Loukissas.
It was also announced this week that Vahdat is one of only 41 researchers worldwide to be awarded a newly-created Hewlett-Packard Labs Innovation Research Award. The award will allow Vahdat and his team to develop further their proposed new networking architecture outlined in their SIGCOMM paper.
The researchers' work addresses problems inherent to current datacenter networks found in scientific computing, financial analysis, social networking, or any industry with large-scale computation or storage needs. Explained Vahdat: "Our work addresses the problem of datacenter network connectivity in a world where consolidation is increasingly taking place in datacenters."
Typically, computers are connected by a network architecture that consists of a "tree" of routing and switching elements regulated by specialized equipment, with expensive, non-commodity switches at the top of the hierarchy. But even with the highest-end IP switches and routers, the networks can only support a small fraction of the combined bandwidth available to end hosts. This limits the overall cluster size, while still incurring considerable costs. Application design is further complicated by non-uniform bandwidth among datacenter nodes, which limits overall system performance.
The UC San Diego researchers' envision creating a datacenter that will have scalable interconnection bandwidth, making it possible for an arbitrary host in the datacenter to communicate with any other host in the network at the full bandwidth of its local network interface. Their approach requires no modifications to the end-host network interface, operating system or applications, and is fully backward compatible with Ethernet, IP and TCP. Ideally, the datacenter would also use inexpensive, off-the-shelf Ethernet switches as the basis for large-scale datacenter networks, thereby replacing high-end switches in much the way that commodity personal computers have displaced supercomputers for high-end computing environments.
"The history of computing and technology has an innumerable set of examples where commodity parts take over the functionality of more specialized pieces of equipment," said Vahdat. "But that commoditization hasn't taken place on the communications side. So we do have these specialized components still living in the network infrastructure that incur significant costs and complexity."
"The other issue is that people are treating the datacenter as a mini-Internet," he continued. "That's fine, but they are then forced to use the same components they might use in the wide-area network with a bunch of concerns that don't come up in the datacenter environment: adversarial environments, attacks, parties that might not trust each other, etc. So you have a lot of functionalities in these specialized switching components that you really don't need in your datacenter, and you wind up paying for it in dollars and in complexity."
Page: 1 of 2(Digg, Technorati, more)
New Paper: Parallel Computing Without Parallel Programming
Learn how domain experts can run VHLL programs like MATLAB® on a variety of high-performance platforms without low-level reprogramming and how to work with the largest datasets and complex algorithms without sacrificing ease of use or reducing productivity.
Spider, the world's biggest Lustre-based, centerwide file system, has been fully tested to support Oak Ridge National Laboratory's new petascale Cray XT4/XT5 Jaguar supercomputer and is now offering early access to scientists.
Read More...
Wolfram Alpha, the Web-based computational engine introduced in May, is not a traditional supercomputing application, but relies on supercomputers to satisfy its unique requirements.
Read More...
There was a new energy at this year's TeraGrid '09 conference thanks to an outstanding turnout for the student program. Thanks to support from the National Science Foundation, more than 100 high school, undergraduate and graduate students were able to participate in the conference.
Read More...
Jul 09 | Engineer Live | The demand for computational tools to underpin the 3D seismic interpretation process has never been more apparent. Read more...
Jul 08 | EE Times | Unemployment for U.S. engineers has reached record levels, according to government figures. Read more...
Jul 08 | Network World | Global spending for 2009 projected to drop 6 percent, for a total of $3.2 trillion. Read more...
Jul 08 | Linux Magazine | Portability or efficiency? Neither is guaranteed when writing explicit parallel code. Read more...
Jul 07 | Ars Technica | Japanese company builds custom ASIC to accelerate real-time ray traced rendering for the auto industry. Read more...
Jul 10 | | Engineers, scientists, and other domain experts depend on the productivity enabled by very high-level language (VHLL) tools like MATLAB® and Python. However, as datasets grow larger and programs get more sophisticated, ordinary desktop computers can no longer keep up. The paper explores how to run VHLL programs on high-performance platforms without low-level reprogramming. Work with large datasets and complex algorithms without sacrificing ease of use or reducing productivity.
Apr 14 | | Many HPC IT departments are feeling the rising pressure to deliver more capacity computing and performance while trying to reduce the total cost of ownership. This white paper discusses how an environmentally-friendly and open-standards HPC building block based computing system using flexible interconnect options helps address capacity computing needs.
Source: Addison Snell, GM/VP, Tabor Research; sponsored by Dell
Many organizations that could benefit from the use of HPC clusters find that it is complicated to get the systems up and running because of limited IT resources or the complexities of the clusters themselves. Learn how the Intel Cluster Ready program, for which Dell was an original partner, seeks to address this challenge for entry level and mid-range HPC users.
BlueArc's Titan architecture represents an evolutionary step in file servers by creating a hardware-based file system that can scale bandwidth, IOPS, and overall data capacity well beyond conventional software-based devices. With its ability to virtualize a massive storage pool of up to four usable petabytes of tiered storage, Titan can scale with growing data requirements, offering a competitive advantage for businesses, researchers, or other enterprises seeking to better manage data growth while still ensuring optimal performance.
Sun Studio Compilers and Tools and Sun HPC ClusterTools allow you to create high performance parallel applications for OpenSolaris, Solaris and Linux. Sun Studio Express 11/08 includes MPI performance analysis capabilities and full OpenMP 3.0 compiler support. Learn about all this and the latest in Sun HPC ClusterTools 8.1.