Rob Thorpe <robert.thorpe@antenova.com> wrote:
+---------------
| I think IBM have spent rather less writing code than they make out.
...
| Only a small portion of the total was on writing code.
| That said, they've still spent an enormous amount on it, as have SGI.
+---------------

Most of what SGI spent on Linux *code* [as opposed to the PR & support
stuff you mentioned for IBM] was in *porting* existing code -- or at
least, the tricky bits of the algorithms -- from Irix, that they had
already spent 100's of man-years on, especially the fine-grained multi-
processor locking stuff that lets the operating system scale somewhat
reasonably to 100's or 1000+ CPUs, e.g.:

    http://biz.yahoo.com/prnews/060815/sftu043.html?.v=67
    SGI Brings Real-Time Performance and Streamlined Cluster
    Management to Open Source Linux With SGI ProPack 5
    Tuesday August 15 [2006]
    ...
    The latest release of SGI ProPack brings SGI's REACT Real-Time
    Extension for Linux to a standard Linux distribution for the
    first time enabling real- time capabilities and support on an
    unmodified version of the open-source kernel. REACT guarantees
    interrupt response times of no more than 30 microseconds in
    configurations of 2 to 64 Intel Itanium 2 processors.

"REACT" existed on Irix years ago.

    SGI ProPack 5, in conjunction with SUSE Linux Enterprise Server
    10, provides another industry first: support for systems running
    1,024 processors under a single copy of Linux. By driving the
    capabilities of SGI Altix and 64- bit Linux to record heights,
    SGI continues to push the boundaries of what even the most
    demanding high-performance computing (HPC) customers can expect
    from Linux.

But here again, they were doing this with *Irix* [on MIPS CPUs]
almost a decade ago.


-Rob

-----
Rob Warnock			<rpw3@rpw3.org>
627 26th Avenue			<URL:http://rpw3.org/>
San Mateo, CA 94403		(650)572-2607