COMMERCIAL NEWS
San Diego, CALIF. — Manish Parks reports that with Turbolinux Inc.’s EnFuzion, it’s no longer computing for the masses but supercomputing by the masses.
The software will enable small to midsize companies to efficiently harvest wasted workstation power by clustering multiplatform networked PCs, workstations and servers, thus attaining supercomputer-level power for performing computationally intensive calculations over multiple CPUs.
eWeek Labs’ tests showed that EnFuzion’s ability to create supercomputing clusters will be invaluable for corporate settings with home-developed, computationally heavy and intensive applications.
The software, which began shipping in July, is flexible enough to accommodate the latency inherent in heavy-duty calculations. Problems can be parallelized, and high network latency does not affect calculations.
We also found no other company that offers such a complete package. Lots of vendors offer hardware and software that provide the basics of superclustering, but nothing is comprehensive enough to compare directly with Enfuzion.
However, even with the potential for unlimited node scalability, at $400 per node, the cost of implementing the current version of EnFuzion could be prohibitive for some small-scale operations.
The potential for misuse of EnFuzion is also great because it provides an efficient tool for cracking passwords.
We strongly recommend that administrators delete the original plain text file upon installation and scramble the standard text password file with the enfdispatcher -e command option. Doing so will prevent most garden-variety hacks into the EnFuzion account, although it’s still vulnerable to sophisticated breaches.
In tests, EnFuzion performed well on multiple platforms, including Microsoft Corp.’s Windows NT Workstation 4.0 and Windows NT Server 4.0 and TurboLinux Inc.’s TurboLinux Workstation 6.0. However, EnFuzion doesn’t support the Windows 9x family, nor does it have a secure-shell implementation.
Lack of Windows 9x support probably won’t seriously crimp EnFuzion’s popularity, and a secure-shell-enhanced version will be released in the near future, as will a version with Windows 2000 support. TurboLinux will include Windows 9x in future releases if there is demand for it, company officials said.
EnFuzion can easily be deployed over the network from a root node, but it cannot be deployed from a Linux node to an NT machine or vice versa.
With recent increases in network bandwidth and CPU speed, paired with a meager average desktop bandwidth utilization rate of approximately 15 percent, the stage is set for supercomputing to come into play in a big way in areas such as gene mapping and other bioinformatics, computer graphics, physics, and very large-scale inte grated circuit design.
Because of their inherent stability and scalability, clusters present a viable alternative to supercomputers for smaller firms.
As market prices for hardware such as memory, hard drives and CPUs have fallen, it has become much easier and less expensive to upgrade a workstation or to set up workstations for redundancy, rather than refit a conventional supercomputer.
Out of the box, EnFuzion is straightforward to use and configure because it utilizes the open-source Tcl (Tool Command Language)/Tk (GUI tool kit) for plan file scripting and GUI displays.
The next release will utilize Java, harnessing its potent scripting abilities to further automate supercomputing clusters. Sites that place a priority on Java support should wait for the next release.
With minor scripting, we could set up options such as recruitment of nodes when in screensaver mode, logged-out mode or time-scheduled node recruitment. These options enable administrators to minimize the effects of supercomputing on daily operations while optimizing system utilization.
In eWeek Labs’ tests, EnFuzion provided near-linear scalability in renderings with POV-Ray, a program that tests supercomputers’ performance by generating or rendering animations by doing hundreds of calculations in parallel.
Nevertheless, scalability is highly dependent on issues such as the number of CPUs involved, network overhead and bandwidth, the type of computations involved, and physical and logical memory.
The testbed consisted of three clustered Dell Computer Corp. Pentium Pro 200-based systems with 128MB of RAM and two 450MHz Compaq Computer Corp. Deskpro EN machines with 192MB of RAM. A resource-intensive animation program running on one Pentium Pro took approximately 22 minutes but only 3 minutes when run on all five machines.
When we took one or more of the machines offline, EnFuzion automatically redistributed the jobs to the other nodes. Visit http://www.turbolinux.com for more information.
============================================================