agoric start performance investigation

0-README.md

context: wallet slow to deploy: 22 minutes · Issue #3802 · Agoric/agoric-sdk

Findings

• onBind look like they could go in parallel
• E(zoe).startInstance() could take an ERef<Instance>
• wallet API is polling timer

Methods

my agoric install comes from 3acf78d786fe from Fri Oct 8

I gathered a sim chain slog file:

date -u && SOLO_SLOGFILE=$PWD/yslow3.slog SLOGFILE=$PWD/simchain.slog agoric start -v --reset

size etc.:

connolly@jambox:~/projects/agoric/testnet-load-generator$ ls -l *.slog
-rw-rw-r-- 1 connolly connolly 93903002 Oct 11 19:25 simchain.slog
-rw-rw-r-- 1 connolly connolly 72561925 Oct 11 19:25 yslow3.slog
-rw-rw-r-- 1 connolly connolly 32152840 Oct 11 19:15 yslow-cont.slog
-rw-rw-r-- 1 connolly connolly 40155861 Oct 11 15:50 yslow.slog

then build a plantuml sequence diagram and render as SVG:

Compilation started at Mon Oct 11 19:25:57

make view
node scripts/slog-to-diagram.mjs <simchain.slog >wallet-start.txt
{ ref: 'kp40' }
...
{ ref: 'kp696' }
java -jar plantuml.jar -tsvg wallet-start.txt
firefox slog.svg

slog.svg

wallet-start.txt

darn it; I clobbered simchain.slog

I love this! Great start!

Looking at this, can you tell where promise pipelining occurred? Where there were forced round trips? How?

Thanks, @erights
A pattern of
->
->
<-
<-

between 2 vats suggests pipelining, or at least no await.

->
<-
->
<-

... suggests an await. Looking at the method names and object identifiers can suggest more strongly that the round trip is unnecessary.