> > We've tested MkLinux on the Compaq SystemPro and on a DEC
> > Celebris (2 Pentiums, using a microkernel binary from the other
> > branch). It works fine.
> When you say "it works fine" does this mean "it works but doesn't
> utilise the 2nd CPU" or does it mean "it works and it utilises the 2nd
> CPU", and if so, are there any benchmarks/numbers to compare how they
> fare in relation to single CPU boards?
> Thanks for your info!!
It works and uses all the CPUs available (Mach works on the
4-pentium AT&T Globalyst S40 machines as well, using the 4 CPUs).
However, Linux has not yet been parallelized, so you can't execute
system calls in parallel. The Linux server is serialized by a
global lock. User tasks can run in parallel but only one system
call can run at a time. I think the Linux/SMP team is working
on parallelizing Linux.
We haven't made any measurement of MkLinux on MP machines yet,
but Mach has proven to be quite MP-efficient and scalable
with OSF/1 (which is fully parallelized). From measurements that
we did a long time ago, it appeared that with 2 cpus, you don't
get much extra speed from fine-grain locks (the overhead of having
to take and release lots of locks is bigger than the gain in
parallelism). It all depends on the kind of benchmark of course.
[ Francois ]
| 2 avenue de Vignate | | phone : (+33) 220.127.116.11 |
| 38610 GIERES FRANCE | | fax : (+33) 76.51.05.32 |