A Cluster by Any Other Name Is Not a Grid
- By John K. Waters
Have you noticed how lately, when the conversation turns to distributed computing,
grid and cluster are used as synonyms? These two words do have different meanings,
Donald Becker reminds us.
“Clustering takes multiple independent machines, and through software
and networking, makes them appear as a single, more powerful and/or reliable
machine,” Becker says. “What you are trying to do is to create a
unified virtual system over multiple single machines. Grid is a concept that
involves working with a large number of separately administered machines. With
grid, you don’t control the configuration, the operating systems, the
libraries installed—anything. So it’s a lot more difficult to ensure
that that application is going to run remotely.”
So-called grid computing solutions for small to midsize businesses are more
likely to be utility computing or clustering solutions, Becker says. A grid
implementation is much more difficult. The only place for "true grid,"
he says, is with global enterprises that have a single point of administration
for the CIO.
Becker knows whereof he speaks. While working at NASA's Goddard Space Flight
Center during the late 1980s, Becker created the first Beowulf cluster, using
software to connect many inexpensive PCs to solve complex math problems typically
reserved for classic supercomputers.
Beowulf is a trademark controlled by Linux International. It defines a type
of scalable performance cluster based on commodity hardware, open-source infrastructure
software and a private system network.
Becker founded Scyld Computing in 1998 to design a new generation of cluster
control software. He now serves as the company's chief scientist. Scyld, which
is a subsidiary of Penguin Computing, develops and sells a cluster operating
system called the Scyld Beowulf Series 29 Linux Clustering Software. Better
known as Scyld Beowulf, the platform is a standards-based Linux distribution
that has been extended through the Scyld BeoMaster technology for ease of deployment
and manageability of Linux clusters. A full high-performance computing software
toolset is integrated with the platform, and it utilizes the Linux Kernel Version
2.4. Most MPI applications running on Red Hat Enterprise Linux ES 3 run unchanged
in the Scyld Beowulf environment, Becker says.
One of the keys to the success of his company’s technology, Becker says,
is the open-source infrastructure on which it runs. "We have some new ideas on how to make large sets of machines very simple to manage," he says.
“But without an open-source infrastructure such as Linux, we wouldn’t
be able to do that.”
Becker finds it "curious" that grid and cluster have become almost
interchangeable terms. "A few years ago, clusters meant simple one-to-one
failover," he says. "The terms have moved from that simple concept
of a single spare taking over a workload to being synonymous with chains spread
across the planet. It’s surprising.”
Becker literally wrote—or rather co-wrote—the book on Beowulf Clustering:
How to Build a Beowulf: A Guide to the Implementation and Application of PC
Clusters (with Thomas Sterling, John Salmon, and Daniel F. Savarese, The
MIT Press, May 28, 1999).
John K. Waters is a freelance writer based in Silicon Valley. He can be reached
at [email protected].