Benchmark hot dapp with real node environment

[aionbot]

Issue created by yulongaion (on Tuesday Sep 04, 2018 at 20:14 GMT)

To get the single-node performance

[aionbot]

Comment by jeff-aion (on Thursday Oct 11, 2018 at 14:47 GMT)

Pushing this to Beta since that is the first time we will have a quasi-real "node environment".

[aionbot]

Comment by jeff-aion (on Monday Nov 19, 2018 at 14:44 GMT)

This should be done early within the Beta timeframe but doesn't block the Beta release.

[aionbot]

Comment by jeff-aion (on Thursday Nov 29, 2018 at 19:45 GMT)

This would do a lot to help us answer the constant question of what our performance is like, would give us an interesting public talking point, and could be used as a coarse guideline by which we could measure optimization work.

I suggest that we create 2 test DApps:

• CPU-heavy
• IO-heavy

From there, create 1 deployment script which will deploy 1000 copies of either one, along with corresponding user identities to call them (so we don't synthetically bottleneck or need to fake odd nonces from one user).

Finally, we create 2 test scripts which can run on either one of these deployed DApp sets:

• saturate many users calling one instance
• saturate many users calling their respective instances

This will allow us to see the impact of hot DApp JIT optimization as well as measure the behaviour of the concurrent executor in a quasi-real environment (real DB, etc).