Talk of the Lab - Farewell, FNALD by Mike Perricone
Farewell, FNALD
Parting is, indeed, such sweet sorrow, although Shakespeare could not have imagined his words applying to a computing cluster.
“So much of the work that people do here is tied up in these systems, that it really is hard to let them go,” said Steve Wolbers, deputy head of Fermilab’s Computing Division, who faces the task of “de-commissioning” the cluster called FNALD by the end of March.
That’s a polite way of saying shutting it down and moving it out, because the Feynman Computing Center needs the space, and needs the power and cooling capacities that FNALD is absorbing.
But computing systems often seem to be obsolete almost as soon as they hit the market. What’s the pull with FNALD? Simply this: in a world of instant change, FNALD has served Fermilab scientists as reliably as an old friend for more than 20 years. Even the “migration” (moving users to different systems) has lasted years longer than expected: begun in 1995, it was supposed to be complete by 1998.
The original Digital Equipment Corporation VAX components and their successors have been in one location or another, in one form or another, since before 1980—before the Feynman Computing Center was constructed, even before CDF itself was completed. First, they served almost the entire Fermilab community, then they stood as the primary computing resource for the CDF collaboration starting in the mid-80s. Why did the system last so long?
“The VMS operating system was extremely efficient in doing scientific computing,” Wolbers said. “The hardware was robust and reliable, the clustering of components meant you add or remove components as needed, it was a common language available in almost every university physics department around the country. It was good. It ran, and ran well, for a long time.”
Installing the VMS clusters represented a switch from mainframe computers to what were called mini-computers at the time. The stacking of components might sound similar to the current use of PC farms, but the VMS cluster offered an even more powerful concept that farms still have not attained: a single file view with single log-in and common access to disc and tape. VMS also offered an advantage over contemporary systems.
“One of the big fights in those days, if you can believe it, was whether or not it was ‘sensible’ to spend money on computers to do ‘chit-chat’ as opposed to only batch computing,” said David Ritchie of CD-Online and Database Systems, who was involved in the initial installation. “It meant doing ‘interactive computing’ in which human beings sit at a terminal and type text into the computer that is to be read somewhere else by another human being, not a computer program. The alternative was writing and running programs whose only purpose was to read data files and write other data file results, which one then transcribed via a typewriter into a paper document submitted for publication.
“What happened,” Ritchie continued, “was that people found that working on the VAX’s was actually quite productive. They had tools like e-mail for communicating with collaborators, good interactive editors, the ability to run really large programs, and an ability to have more than one person active on the computer at a time. Thus, people quickly gravitated to it as a place to work and to collaborate.”
Importantly, individual collaborations could add components, as, for example, CDF’s Italian contingent did around 1986 (as shown in the photo, with computer department head Hugh Montgomery, left, and research head Ken Stanfield, right, flanking CDF collaborators Antonio Bellini and Giorgio Bellettini).
“In its heyday, FNALD had a total of 21 computers, four of which I know for sure were supplied by the Italians,” said Rick Colombo of CD-CDF Computing and Analysis, who has maintained FNALD through much of its lifetime. “Those four computers were part of 11 computers that we called the ‘K’ boxes. They were used within FNALD much like we now use a farm, only they were literally part of FNALD and not isolated. Certain jobs were run specifically on these 11 nodes, the Italians getting some preferential treatment running batch jobs on the ‘K’ boxes for their contribution.”
But finally, the PC explosion, the advent of Windows and LINUX, and the direction of the computing industry meant the bell was tolling. The Computing Division will hold a send-off on March 27.
“It did a great job in its time,” Wolbers said. “It’s time to move on.”
|
last modified 3/15/2002 email Fermilab |
FRLsDFx9eyfrPXgV