johansen wrote:
With Solaris 10, Sun undertook a very specific set of projects with the explicit objective of improving the performance of small (1-2p) systems. I can't speak for Phil, but in my experience performance engineers have always chosen the solution which scales best, since we have to support 1-128 processors with the same kernel. The implication here, is that although we scale well, the cost for 1-2p may be unexceptably high. Part of the small-systems work that we undertook, and that Bart championed, was to find these cases where we out-scaled Linux but performed worse on small systems or small workloads. In these situations a lot of the engineering work involved figuring out a compromise which allowed us to scale well on large machines, but to eliminate the overhead on small machines. Clearly, there is still work to do, and Linux is a moving target.
Indeed, but we goad ourselves with slogans such as "If Linux is faster, it's a Solaris bug" ...
http://blogs.sun.com/roller/page/pgdh?entry=if_linux_is_faster_itOf course, there are always exceptions (sometimes scalability matters most). But the real challenge (and the most fun) is to produce something which is fast on 1P/2P systems and which scales well on large multiprocessor configurations too. For instance I had a whale of a time making getenv(3C) both fast and scalable ...
http://blogs.sun.com/roller/page/pgdh?entry=caring_for_the_environment_making
We've made a good progress. Today, Solaris is a lot more competitive with Linux on 1P/2P systems (including plenty of stuff which we are actually better at). But there is still plenty of scope for lots more fun! And with OpenSolaris we're saying we don't want to keep all the fun to ourselves :)
Phil http://blogs.sun.com/pgdh
smime.p7s
Description: S/MIME Cryptographic Signature
_______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
