Implement performance profiling

[vukolic]

We need to implement performance profiling to understand (at any point in time) where the performance bottleneck in openblockchain is.

For example, at the time of opening this issue we are at about 500 tx/s peak throughput - which is way too low - yet it is not easy to tell where the bottleneck is.

Besides being able to identify throughput bottlenecks, we need also latency profiling capabilities.

[corecode]

With transaction batching, I can get to 600 invokes/sec. My feeling is (given that the profiler does not show any big offenders) that the big speedbump is transaction execution latency to the chaincode and back.

[muralisrini]

adding @dongmingh @bmos299... they have started profile work

[corecode]

As suggested by @muralisrini, I tested with a noop chaincode, i.e. Run() and Query() return immediately (but the shim code is still in place). I can get up to 2500 invokes/sec on my laptop, and 2200 invokes/sec on zc2.

[vukolic]

@corecode is this the throughput of a single client in closed loop or w. multiple clients saturating the system?

[corecode]

system saturated, closed loop clients.

[vukolic]

Thanks. This number is very low on its own in the first place (although the rest of the system is not there yet).