Memory system call such as mmap, malloc, POPULATE etc... shows slow performance..
The root cause is very similar to the one with the file system call performance..
- kernel data structrue, such as mmstruct, vmastruct, are shared across the GiantVM on top of the DSM layer, and the DSM layer performance is not sufficient for this task..
- Lots of synchronization is also dependent on the slow DSM layer.
Our experimental were done with our own ad-hoc microbench, called memcpy.c
More good microbench for this purpose is a necessary.
Either find good memory benchmark or develop one...
Memory system call such as mmap, malloc, POPULATE etc... shows slow performance..
The root cause is very similar to the one with the file system call performance..
Our experimental were done with our own ad-hoc microbench, called memcpy.c
More good microbench for this purpose is a necessary.
Either find good memory benchmark or develop one...