Closed jcnelson closed 9 years ago
Actually, what we're really after is a global limit on number of cached blocks. We can track this through the cache by having the cache give us back allocated dirty blocks.
EDIT: implicitly, we're fine as long as the number of open files is kept under control.
We've addressed this in the new UG's block subsystem. Each file inode has at most one cached dirty block in RAM, and blocks on disk will be mmap(2)'ed as needed.
See if we can do away with the UG's dirty blocks subsystem, and put this directly into the cache. What are the implications for doing so?