Voir en

français

25 years ago: the strategic move to PCs in high-energy physics

In September 1995, a paper entitled “The PC as Physics Computer for LHC?” paved the way to a new era in high-energy physics computing

The new Computer Centre
The new Computer Centre in building 513. (Image: CERN)

When the Large Electron-Positron Collider (LEP), the LHC’s predecessor, began running in 1989 the central computing facilities were almost entirely based on mainframes and supercomputers. These computers offered reliable services but were very expensive so that the installed capacity was rather limited.

Due to the high costs, the CERN IT department (then “CN”) and the LEP physicists started turning to RISC (reduced instruction set computer) workstations and servers running UNIX. These computers were mainly targeting interactive users, but when used for batch processing they offered a much more attractive price/performance.

Having observed this advantageous move to RISC, CERN decided to investigate if the same “trick” could be done with PCs and started a batch porting and benchmarking project inside the RD47 collaboration.

PCs had become prevalent as office and home computers but were not really targeting “serious” batch computing.

Fortunately, for CERN, the Intel Pentium processors which were announced in 1993, had much better floating-point support than the previous generations and High-Energy Physics (HEP) programs need good floating-point performance.

But even with improved hardware, the porting and benchmarking turned out to be rather difficult. CERN had to choose from multiple operating systems and a large number of compilers and each environment offered challenges.

By the summer of 1995, we had finally obtained encouraging results and decided to present at CHEP-95, the 8th International Conference on Computing in High Energy and Nuclear Physics, held in Rio de Janeiro in the week of September 18-22. Our paper was entitled “The PC as Physics Computer for LHC?” and was well received. The decisive argument was, as it had been for RISC servers, that the price/performance was drastically improved.

After CHEP, we presented the results in several HEP institutes, such as FNAL, SLAC and DESY. One thing, however, needed to be corrected. We had chosen Windows/NT as the operating system because it supported dual-processor PCs, allowing the best price/performance, but the physicists greatly preferred UNIX/Linux. This was solved a year later when Linux 2.0 added SMP support.  With this combination of hardware and software the PCs quickly conquered all the HEP computing centres and quite amazingly, PCs have remained practically unchallenged ever since. HEP has, of course, not driven this evolution, but has piggybacked on the fact that “everybody” decided to move to PCs over the years.

It is hard to predict what the next 25 years will bring but PC technology is, by no means, ready to leave the scene.

Sverre Jarp

 

How the CERN Data Centre has evolved over the years. (Video: CERN)