Dennis Clarke wrote:
> Sequential consistency and instruction order issues across threads is very
> non-trivial.
>
> Millions of dollars of R&D required.
Which, by the way, IBM provided for Linux.
They had several developers working for well over a year with the Linux folks
on
the port to POWER6, and even the IBM engineers had a difficult time getting the
memory ordering code right.
Lots of stupid little bugs needing synchronization in non-obvious places that
took months to track down each time. :-(
Now given that IBM sees a business case in getting people off Sun hardware onto
their mainframes and is thus paying SNA to port OpenSolaris to Z, one MIGHT be
able to make the argument that they might similarly be able to justify a
business case in helping a group port OpenSolaris to POWER6, but that's BEYOND
even wild speculation on my part.
William Kucharski
