perf(channel): performance improvements

[ramey]

Changes

The following changes are done as part of this PR -

• Remove serde interfaces to use direct functions
• Fix shutdown
• Change ProduceBulk method to remove CollectRequest

Closing #35

[NNcrawler]

The PR is LGTM. Can you also put the benchmark result as well in the issue?

[ramey]

Performace Benchmarking

Setup

Max Clients - 1000
Batch Size - 100
Batch Interval - 10000
Execution time - 30m

v0.1.1

Server Processing Latency(max_upper)

Min 624ms
Max - 6.71 min
Avg - 45.8s

Batch idle time in channel

Min - 1.03 min
Max - 7.6 min
Avg - 5.6 min

Worker Latency(max_upper)

Min 191 ms
Max 1.02 min
Mean 8.92s

v0.1.3

Server Processing Latency(max_upper)

Min - 6.74s
Max - 17.9 min
Avg 7.79 min

Batch idle time in channel

Min - 2.20 min
Max - 9.09 min
Avg - 6.54 min

Worker Latency(max_upper)

Min - 100ms
Max - 1 min
Avg - 13.2 s

Current Branch

Server Processing Latency(max_upper)

Min - 15ms
Max - 7.04 min
Avg - 1min

Batch idle time in channel

Min - 1.63s
Max - 7.08 min
Avg - 3.69 min

Worker Latency(max_upper)

Min 58ms
Max 3.04s
Avg 512 ms