We have proposed an ``Asymmetric Distributed Shared Memory: ADSM'', that
realizes user-level protected high-speed
communications/synchronizations.
In the ADSM, the shared-read is based on a cache-based shared virtual
memory system. As for the shared-write, instructions for consistency
management are inserted after the corresponding store instruction.
Therefore, various optimizations can be performed. 
We propose an optimizing method of reducing overheads for consistency
management. The algorithm coalesces a sequence of consistency
management instructions statically/dynamically.
We have implemented the prototype of the compiler and the runtime system
for the ADSM on a multicomputer Fujitsu AP1000+ and the
general-purpose massively-parallel operating system: SSS-CORE. 
The performance evaluation using LU-Contig of SPLASH-2 shows that 
the execution time is reduced by 80% using static optimization and
it is further reduced by 30% using dynamic optimization.