eth/tracers: improve tracing performance

[holiman]

This PR first adds a benchmark of the "opcode tracer", i.e, the barebones tracer which stores data in memory before sending it out over RPC. The benchmark is a very simple loop which just runs through 1M gas.

The second commit adds some optimizations which improves the performance quite a lot:

name                old time/op    new time/op    delta
TransactionTrace-6     100ms ±12%      28ms ±13%  -72.03%  (p=0.000 n=10+9)

name                old alloc/op   new alloc/op   delta
TransactionTrace-6    40.1MB ± 4%     9.0MB ±12%  -77.58%  (p=0.000 n=10+10)

name                old allocs/op  new allocs/op  delta
TransactionTrace-6      571k ± 0%      163k ± 0%  -71.42%  (p=0.000 n=10+8)

[holiman]

Further improvements:

name                old time/op    new time/op    delta
TransactionTrace-6    28.1ms ±13%    22.7ms ± 8%  -19.07%  (p=0.000 n=9+9)

name                old alloc/op   new alloc/op   delta
TransactionTrace-6    8.98MB ±12%    7.14MB ±14%  -20.55%  (p=0.000 n=10+10)

name                old allocs/op  new allocs/op  delta
TransactionTrace-6      163k ± 0%       82k ± 0%  -49.98%  (p=0.002 n=8+10)

Total:

name                old time/op    new time/op    delta
TransactionTrace-6     100ms ±12%      23ms ± 8%  -77.36%  (p=0.000 n=10+9)

name                old alloc/op   new alloc/op   delta
TransactionTrace-6    40.1MB ± 4%     7.1MB ±14%  -82.18%  (p=0.000 n=10+10)

name                old allocs/op  new allocs/op  delta
TransactionTrace-6      571k ± 0%       82k ± 0%  -85.71%  (p=0.000 n=10+10)

[holiman]

I ran this against this transaction,

#!/bin/bash

time ./build/bin/geth attach --exec "debug.traceTransaction('0x87ba373aa69ee675ae764b49c3d179d2367dd0d970c36b9177dc67f00603b4e9', {'disableMemory':true, 'disableStorage':true})" 

It took about 52s, and 27s with stack disabled.
I also took a cpu profile when running it with stack enabled:

[MariusVanDerWijden]

So I tried this on a really heavy tx we have on ropsten
time ./build/bin/geth --ropsten attach --exec "debug.traceTransaction('0xcafaf3669df640365da47527902e3e4dff48bd334ab0c5c894dae5ba1f82df27', {'disableMemory':true, 'disableStorage':true, reexec:10000})"

This PR:
(60 seconds of it is state regeneration)

Executed in  281.69 secs    fish           external
   usr time  151.47 secs  323.00 micros  151.47 secs
   sys time  103.44 secs   67.00 micros  103.44 secs

Current Master:
(59 seconds of it is state regeneration)
(I killed the node, call didn't finish)

Executed in  269.17 mins    fish           external
   usr time    6.21 secs  358.00 micros    6.21 secs
   sys time    0.50 secs   77.00 micros    0.50 secs

[holiman]

So I tried this on a really heavy tx we have on ropsten

Looks like it worked then? Can the PR haz an approve?

[MariusVanDerWijden]

LGTM

[holiman]

0xcafaf3669df640365da47527902e3e4dff48bd334ab0c5c894dae5ba1f82df27

18M gas on that transactions -- yeah might be a bit slow :)