It was like dropping a red handkerchief at a bullring. On March 8, IBM developer Martin Bligh posted notice to the Linux kernel development list that he had compiled the Linux 2.4.18 kernel in 23 seconds.
Bligh used a 16-CPU IBM NUMA-Q box with 4GB of RAM and had whittled that time down from his previous best of 47 seconds through a series of patches tuning Linux for NUMA-Q hardware.
Well, five days later, Anton Blanchard, also at IBM but working on PowerPC hardware, knocked off that number quite handily with a 10.31-second compile of the same kernel using 24 1.1GHz 64-bit Power4 CPUs and 60GB of RAM.
Blanchard didnt identify the exact machine, but the specs match those of a big IBM pSeries 690 box, which normally runs AIX 5.1 and goes through a million dollars of capital expenditure without breathing hard. Its very interesting to see that IBM has people working on tuning Linux performance on its top-of-the-line Unix hardware.
That wasnt the end of it, though. Three days after his first post, on March 16, Blanchard got his kernel compile down to 7.52 seconds by adding eight more CPUs and rewriting the Power-PC 64 memory page-table management code for better performance. “I think Im addicted. I need help!” he wrote.
That record stands—for now.