With availability-enabled set to true running the trade2 benchmark with
profiling info being collected, it was identified that GMS deserialization was
taking a substantial amount of time.
Given the buffering of replicated data that HA is using, it is believed that the
message size is around 80K. Given that GMS over Grizzly transport is relying on
composite buffer (virtual buffer that spans multiple grizzly physical buffers),
there is potential that improvements to that would assist this issue.
Alexey will assist in improving composite buffer management. GMS team will also
work on stats monitor so we know precisely what the usage characteristics are
and to track where the time is going during deserialization in this benchmark.