Stephen D. Brown
- Reports describing various
aspects of the NUMAchine project can be obtained via the WWW
- Although the first working
NUMAchine implementation will use the MIPS R4400 processor, decoupling
the processors and memory in our implementation of the NUMAchine
architecture permits switching to the MIPS R10000 processor when it
- A single queue would suffice if one could
guarantee that a nonsinkable message is received only when it can be
fully processed without blocking subsequent sinkable messages.
- Atomic access is performed using
read_with_lock and write_with_unlock operations.
The lock acquired is the per-cache line lock used by the coherence
protocol, allowing system software to manipulate cache lines even
while they are being accessed by application software.
- The SPLASH-2 applications take
anywhere from 5 minutes to 24 hours to run on an SGI Challenge for each
different system configuration. These times can double if both effects are
- For the sake of other SPLASH-2 experimenters,
the BlockOwner routine was changed to
Wed Jun 28 18:31:42 EDT 1995