I saw Rick Lindsley of the IBM Linux Technology Center (LTC) at a lecture
last December. He had been working for Sequent for 9 years when IBM bought
Sequent, formed the Linux Technology Center with the formerly Sequent
programmers as the core and has turned them loose on Linux kernel
development.
The LTC has rewritten the scheduler in the 2.4 kernel and improved
performance by 25% on a 4 processor computer to where it is 2.5 times the
speed of a single processor computer. They are confident that they should
be
able to achieve 3.99 times the speed of a single processor. They have
already done this on Sequent's own OS. Their plans for the future include
support for 16 processor computers. All of this without having an effect on
single processor performance (one of Linus' personal requirements).
The first step was to do a major rewrite of the scheduler and then prove to
Linus that they were still using the same algorithm, just cleaner code.
Then
they started doing multiprocessor improvements. Alan Cox has signed on to
do
peer review of their work. Also they want to do other work on hardening the
kernel and file system redundancy. It is IBM's view that if Linux is to
reach common use in big iron, it must be much more dependable than it is
currently. That is IBM's goal.
Rick said that currently the LTC has around 40 programmers. They plan on
expanding to around 80 to 90 programmers, not including support staff. The
facility is located in Beaverton, Oregon, though many of the LTC personnel
are located around the world.
IBM believes that the importance of Linux is only going to grow. As such it
is better to get involved now than to have to play catch up later. They are
willing to play the game by the OSS rules.
</x-flowed>
