#
Chocolate liegeois
#
OIT needs help; University must find, grant resources**
If you see the letters "rgb" and see red, if you see the phrase "On_spin3d" and grow angrier still, then you know why I'm writing this column. Throughout last semester and the beginning of this one, physics professor Robert Brown has used the entire acpub system as his own supercomputer. On all 180 plus of the SUN workstations installed by the Office of Information Technology (OIT) last year, Brown continuously runs his program "On_spin3d." The effect of his work is to reduce by half the speed with which each of the computers performs mathematically intensive functions.
OIT purchased a fantastic computer system last year of which students and faculty only use a fraction of the available computing power. Why should a valuable resource go wasted? The problem, of course, is that in using acpub to its capacity, Brown has inconvenienced other users.
When Brown started his program, he made every effort not to disturb others' use of the system with a utility called "nice." With "nice" operating normally, Brown's program would not affect anyone else's use of acpub. Unfortunately, the scheduler (called a dispatch table), which permits "nice" to function in a useful manner, is not configured correctly. Thus, although Brown is trying to be nice, the effort has not succeeded.
Should we blame Brown for the problem? No. Although inconveniencing other cluster users is not a nice thing to do, letting excess computing power go to waste would serve no one. According to Dr. Brown, he has told OIT on several occasions how to fix the problem and even offered to implement the solution, but as yet he has received no response.
Of course, many will say, "Who cares?" OIT has done a wonderful job maintaining a complicated network with a small, overworked staff performing miracles, handling an enormous increase in computer usage this past semester. Why should they bother fixing a bug that only affects those doing math intensive functions?
Brown's research is legitimate, and now that he has shown acpub's potential to the professorate, what happens next? Will professor after professor line up to use acpub resources to solve their problems? How will OIT decide between equally valid needs? As long as acpub is not "nice," distributional questions about the free use of a supercomputer will surface.
More significantly, the failure to at least respond to a legitimate concern regarding network administration points to a deeper problem. Now that Duke has made a large investment in UNIX workstations, it needs to invest equally heavily in a sufficient support staff to keep acpub tuned to peak efficiency. Knowing something of the difficulty those currently administering the system face, I cannot bring myself to blame them for failing to address what, for them, must be a minor problem. If you consider how many people use computers on Duke's campus everyday and how frustrated people get when their computers do not work as expected, the current OIT staff faces an untenable problem.
The problem lies in not finding a full time, professional UNIX system administrator whose only job is to manage high level acpub computing concerns. The problem with "nice," as well problems with other network issues such as the length of time it takes for remote shells to appear, clearly indicates that the current acpub staff needs more support from their administrative superiors.
Recognizing that the University has just spent a great deal of money to improve computing hardware at Duke and that funds are exceptionally tight (few academic departments received budget increases last year), where will OIT find the money to complement the existing staff with new specialized hires?
If nothing else, Brown has proved that large networks of workstations can outperform supercomputers on certain problems. Moreover, if "nice" were working, the cost to other users would prove minimal. Creative uses of computer power such as this could prove useful to agencies such as the National Science Foundation who routinely give grants for supercomputers in the millions of dollars if the problems requiring such power have sufficient merit. After speaking with Brown, he said that he was unaware of any other university's computing resources being used in the same manner as ours. Why not make an NSF proposal requesting funds to hire more network personnel and to purchase more workstations? The money would go toward researching how to use wasted power on existing networks. Students and professors would benefit by having more computers. OIT would benefit by acquiring more staff. Duke would benefit by being a leader in yet another area.
While creating such proposals takes significant up-front costs in terms of labor and initial research, the benefits of success are clear and compelling.
Alex Rogers is a Trinity senior.
Get The Chronicle straight to your inbox
Signup for our weekly newsletter. Cancel at any time.