This makes is much easier to process on the server and opens new
optimization possibilities. It also fixes theoretical problems, which
may be caused by invalid ordering of events with the same timestamp.
This change exploits the fact that events are processed in batches
originating from a single thread. A single message changing thread
context is enough to handle multiple messages, as opposed to inclusion
of thread identifier in each message.
Compressing frame images on a separate thread may cause frame image
arrival before frames are sent. Fix this issue by creating a staging
area in which frame images will wait for frames to arrive.
This probably breaks playback functionality, as non-existent frames may
be queried, but this problem seems to be very hard to find, so let's
ignore it for now.
Zone children will be now collected in staging vectors. When the zone is
ended (and no children can be added anymore to it), a size-fitted vector
is allocated using slab allocation. The over-allocated vector is then
put into cache for use in future zones.
This is only active for vectors <= 8192 elements, or 64 KB (chosen
arbitrarily), to reduce time spent on copying memory.
Overall, this change should have the following effects:
- System memory allocation pressure reduction, due to re-usage of
vectors, which eliminates the need for constant growth.
- Reduction of memory usage, because children vectors are now fitted to
required size.
- Slight increase of zone processing time, due to memory copying?
