[bind10-dev] shared memory data source: # of zones / shmem segment

[JINMEI Tatuya / 神明達哉]

This is the second of my big questions regarding the shared memory
based in-memory data source: especially if the server maintains a
large number of zones or a few of very large zones, should we have all
of them in the same single shared memory segment?  Or should we allow
maintaining them in multiple segments, each contains a reasonably
small number of zones?

Using a single segment is obviously simpler, but that will require
making a copy of the entire segment (again, either on-disk or
in-memory) even when we only need to update one zone in order to allow
user applications (mainly b10-auth) to keep using the old version
while the "memory manager" is making the new version.  Diving the set
of zones into multiple segments will mitigate this overhead, but it
will require more complexity.  Also note that (as far as I can see) we
can avoid the fully copy by exploiting the concept of copy-on-write.

Right now I'm not sure which approach we should take for the coming
task.  Maybe we begin with a single segment and consider extending it
as we see the need?

---
JINMEI, Tatuya