process: improve performance of nextTick #8932

evanlucas · 2016-10-04T23:34:01Z

Checklist

make -j8 test (UNIX), or vcbuild test nosign (Windows) passes
commit message follows commit guidelines

Affected core subsystem(s)

process

Description of change

This replaces TickObject with an object literal. This offers
performance improvements of up to ~20%.

Results from a few benchmark runs:

                                              improvement significant      p.value
 process/next-tick-breadth-args.js millions=2      9.46 %         *** 1.602162e-15
 process/next-tick-breadth.js millions=2          17.76 %         *** 6.210386e-25
 process/next-tick-depth-args.js millions=2        3.19 %         *** 8.427012e-05
 process/next-tick-depth.js millions=2             1.35 %             5.028960e-02

                                              improvement significant      p.value
 process/next-tick-breadth-args.js millions=2      9.30 %         *** 1.815388e-20
 process/next-tick-breadth.js millions=2          18.25 %         *** 1.806162e-22
 process/next-tick-depth-args.js millions=2        1.57 %             8.013786e-02
 process/next-tick-depth.js millions=2             1.98 %           * 2.573876e-02

                                              improvement significant      p.value
 process/next-tick-breadth-args.js millions=2      8.56 %         *** 1.609244e-18
 process/next-tick-breadth.js millions=2          19.44 %         *** 1.270555e-25
 process/next-tick-depth-args.js millions=2        2.52 %          ** 2.739874e-03
 process/next-tick-depth.js millions=2             2.11 %          ** 1.289425e-03

/cc @trevnorris since you added TickObject

Fishrock123 · 2016-10-04T23:43:49Z

Interesting, there is a few places we moved to constructors for previous optimizations, perhaps we should be moving away now?

evanlucas · 2016-10-04T23:56:04Z

@Fishrock123 possibly. Hopefully, I'll be digging some other areas as well this week. I was planning on looking at immediates next, so I'll probably have some questions for you :]

I also need to test on v6.x and see if there are any differences since we updated to V8 5.1

claudiorodriguez · 2016-10-05T08:03:23Z

lib/internal/process/next_tick.js

@@ -151,7 +145,11 @@ function setupNextTick() {
        args[i - 1] = arguments[i];
    }

-    nextTickQueue.push(new TickObject(callback, args));
+    nextTickQueue.push({
+      callback: callback,


Just out of curiosity, is there a particular reason not to go

nextTickQueue.push({ args, callback, domain: process.domain || null });

?

I just find it less readable than the way it currently is.

Prefer what @claudiorodriguez suggested

assuming it does not alter perf that is

it doesn't appear to make a difference. Just ran again using shorthand and the numbers are more or less the same. :]

AndreasMadsen · 2016-10-05T09:05:28Z

Great work. Nice to see that you are using the new benchmarking tool. However you shouldn't do many independent compare.js runs. If you want more certainty just increase the --runs parameter in compare.js. This is not just for convenience, it's also important from a statistical perspective.

For each comparison you do, there is by default a 5% risk that we see a significant result when there in reality no improvement. By doing 3 runs you increase that risk to 3 * 5% = 15% (mathematical inaccurate). On top of this you are doing 4 comparisons with different parameters (nothing wrong here), which increases that risk even more, but that is another story.

evanlucas · 2016-10-05T10:31:06Z

@AndreasMadsen thanks for the info. I ran the bench again once with 100 runs this time and got very similar results:

                                              improvement significant      p.value
 process/next-tick-breadth-args.js millions=2      9.09 %         *** 1.623609e-39
 process/next-tick-breadth.js millions=2          18.32 %         *** 9.868966e-75
 process/next-tick-depth-args.js millions=2        2.55 %         *** 3.923716e-07
 process/next-tick-depth.js millions=2             2.03 %         *** 1.231088e-06

Should I be doing these runs once for each benchmark script?

claudiorodriguez · 2016-10-05T10:38:54Z

CI: https://ci.nodejs.org/job/node-test-pull-request/4393/

claudiorodriguez

LGTM

Fishrock123 · 2016-10-05T13:44:18Z

I'd like to see some info on what branches this can effectively go to before we land it though :)

evanlucas · 2016-10-05T13:57:35Z

@Fishrock123 yea, am going to bench on v6.x now

jbergstroem · 2016-10-05T13:59:24Z

(pending checks are a bug -- i need to cover skipped status)

evanlucas · 2016-10-05T14:19:41Z

results for v6.x:

                                              improvement significant      p.value
 process/next-tick-breadth-args.js millions=2     15.16 %         *** 3.828618e-92
 process/next-tick-breadth.js millions=2          10.47 %         *** 2.817494e-51
 process/next-tick-depth-args.js millions=2        1.38 %         *** 6.977638e-04
 process/next-tick-depth.js millions=2             1.73 %         *** 6.359614e-04

AndreasMadsen · 2016-10-05T17:42:28Z

@evanlucas Should I be doing these runs once for each benchmark script?

Not sure what you mean. But it looks like you did it correctly 👍

imyller

LGTM

imyller · 2016-10-05T18:37:18Z

Oh, what I meant to say is LATM. Looks Awesome To Me.

jbergstroem · 2016-10-05T19:19:16Z

LGTM

evanlucas · 2016-10-05T21:45:01Z

Running CI one more time since I updated it: https://ci.nodejs.org/job/node-test-pull-request/4403/

lpinca

LGTM

trevnorris · 2016-10-06T10:33:26Z

LGTM. Though we shouldn't put much on the breadth tests. 99% of the time the nextTickQueue array has fewer than 5 callbacks when processed.

jasnell

LGTM

evanlucas · 2016-10-10T21:49:48Z

I am going to benchmark on v4.x tonight to see if it is even worth it there as well.

evanlucas · 2016-10-11T12:52:44Z

So, for v4, I had to use the old benchmarks. It doesn't seem to make as much of a difference as on v6.

running ./node_after_v4
process/next-tick-breadth-args.js
process/next-tick-breadth.js
process/next-tick-depth-args.js
process/next-tick-depth.js
running ./node_before_v4
process/next-tick-breadth-args.js
process/next-tick-breadth.js
process/next-tick-depth-args.js
process/next-tick-depth.js

process/next-tick-breadth-args.js millions=2: ./node_after_v4: 1.9412 ./node_before_v4: 1.7772 . 9.23%
process/next-tick-breadth.js millions=2: ./node_after_v4: 3.1019 ./node_before_v4: 3.1364 ..... -1.10%
process/next-tick-depth-args.js millions=2: ./node_after_v4: 10.658 ./node_before_v4: 10.308 ... 3.40%
process/next-tick-depth.js millions=2: ./node_after_v4: 18.469 ./node_before_v4: 18.68 ........ -1.13%

so, we can probably just not backport this to v4.x

AndreasMadsen · 2016-10-11T13:56:48Z

@evanlucas you are supposed to be able to use the new benchmarking suite on old node versions. The process is:

git checkout upstream/v4.x-staging # or similar branch
make -j8
cp ./node ./node-v4

apply-pr 8932 # fix the file missing conflict
make -j8
cp ./node ./node-v4-pr8932

git checkout master
make -j4 # or just use your globally installed version

./node ./benchmark/compare.js \
  --old ./node-v4 --new ./node-v4-pr8932 \
  --filter next-tick -- process | tee results.csv
cat results.csv | Rscript ./benchmark/compare.R

Unfortunately some new syntax has sneaked into common.js. Replacing ...args with args = Array.prototype.slice.call(arguments, 1); fixes it (easy first contribution anyone!).

This gives the result (30 runs):

                                              improvement significant      p.value
 process/next-tick-breadth-args.js millions=2     12.60 %         *** 4.418516e-28
 process/next-tick-breadth.js millions=2           4.47 %         *** 5.463876e-06
 process/next-tick-depth-args.js millions=2       -1.01 %             4.254800e-01
 process/next-tick-depth.js millions=2             3.30 %           * 2.142562e-02

This replaces TickObject with an object literal. This offers performance improvements of up to ~20%. PR-URL: nodejs#8932 Reviewed-By: Claudio Rodriguez <cjrodr@yahoo.com> Reviewed-By: Roman Reiss <me@silverwind.io> Reviewed-By: Ilkka Myller <ilkka.myller@nodefield.com> Reviewed-By: Luigi Pinca <luigipinca@gmail.com> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Johan Bergström <bugs@bergstroem.nu> Reviewed-By: Trevor Norris <trev.norris@gmail.com>

evanlucas · 2016-10-13T17:18:17Z

Landed in 804d57d. Thanks!

This replaces TickObject with an object literal. This offers performance improvements of up to ~20%. PR-URL: #8932 Reviewed-By: Claudio Rodriguez <cjrodr@yahoo.com> Reviewed-By: Roman Reiss <me@silverwind.io> Reviewed-By: Ilkka Myller <ilkka.myller@nodefield.com> Reviewed-By: Luigi Pinca <luigipinca@gmail.com> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Johan Bergström <bugs@bergstroem.nu> Reviewed-By: Trevor Norris <trev.norris@gmail.com>

Using new syntax such as `...args` means that the benchmark suite can't be used with older node versions. This changes the `...args` syntax to a good old `Array.prototype.slice`. Refs: #8932 (comment) PR-URL: #9064 Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Brian White <mscdex@mscdex.net>

MylesBorins · 2016-11-11T19:29:22Z

@evanlucas how long should we be letting this live on v7 before backporting?

evanlucas · 2016-11-11T19:31:38Z

It's a pretty small change so it doesn't really matter to me. I think it should be fine to get in to the next v6.x release if no one objects

MylesBorins · 2016-12-21T18:15:20Z

This does not land cleanly in LTS v4.x. Added dont-land label. Please feel free to manually backport

This replaces TickObject with an object literal. This offers performance improvements of up to ~20%. PR-URL: #8932 Reviewed-By: Claudio Rodriguez <cjrodr@yahoo.com> Reviewed-By: Roman Reiss <me@silverwind.io> Reviewed-By: Ilkka Myller <ilkka.myller@nodefield.com> Reviewed-By: Luigi Pinca <luigipinca@gmail.com> Reviewed-By: James M Snell <jasnell@gmail.com> Reviewed-By: Johan Bergström <bugs@bergstroem.nu> Reviewed-By: Trevor Norris <trev.norris@gmail.com>

evanlucas · 2016-12-23T22:07:21Z

Backport for v4.x can be found at #10433. Thanks!

This LTS release comes with 312 commits. This includes 229 that are test related, 62 that are docs related, 17 which are build / tools related, and 4 commits which are updates to dependencies. Notable Changes: * build: - shared library support is now working for AIX builds (Stewart Addison) #9675 * deps: - *npm*: upgrade npm to 3.10.10 (Rebecca Turner) #9847 - *V8*: Destructuring of arrow function arguments via computed property no longer throws (Michaël Zasso) #10386) * inspector: - /json/version returns object, not an object wrapped in an array (Ben Noordhuis) #9762 * module: - using --debug-brk and --eval together now works as expected (Kelvin Jin) #8876 * process: - improve performance of nextTick up to 20% (Evan Lucas) #8932 * repl: - the division operator will no longer be accidentally parsed as regex (Teddy Katz) #10103 - improved support for generator functions (Teddy Katz) #9852 * timers: - Re canceling a cancelled timers will no longer throw (Jeremiah Senkpiel) #9685 PR-URL: #10394

This LTS release comes with 312 commits. This includes 229 that are test related, 62 that are docs related, 17 which are build / tools related, and 4 commits which are updates to dependencies. Notable Changes: * build: - shared library support is now working for AIX builds (Stewart Addison) nodejs/node#9675 * deps: - *npm*: upgrade npm to 3.10.10 (Rebecca Turner) nodejs/node#9847 - *V8*: Destructuring of arrow function arguments via computed property no longer throws (Michaël Zasso) nodejs/node#10386) * inspector: - /json/version returns object, not an object wrapped in an array (Ben Noordhuis) nodejs/node#9762 * module: - using --debug-brk and --eval together now works as expected (Kelvin Jin) nodejs/node#8876 * process: - improve performance of nextTick up to 20% (Evan Lucas) nodejs/node#8932 * repl: - the division operator will no longer be accidentally parsed as regex (Teddy Katz) nodejs/node#10103 - improved support for generator functions (Teddy Katz) nodejs/node#9852 * timers: - Re canceling a cancelled timers will no longer throw (Jeremiah Senkpiel) nodejs/node#9685 PR-URL: nodejs/node#10394 Signed-off-by: Ilkka Myller <ilkka.myller@nodefield.com>

evanlucas added process Issues and PRs related to the process subsystem. performance Issues and PRs related to the performance of Node.js. labels Oct 4, 2016

claudiorodriguez reviewed Oct 5, 2016

View reviewed changes

claudiorodriguez approved these changes Oct 5, 2016

View reviewed changes

silverwind approved these changes Oct 5, 2016

View reviewed changes

imyller approved these changes Oct 5, 2016

View reviewed changes

lpinca approved these changes Oct 6, 2016

View reviewed changes

jasnell approved these changes Oct 6, 2016

View reviewed changes

evanlucas force-pushed the nexttickperf branch from 79c679d to fc514fd Compare October 11, 2016 12:53

AndreasMadsen mentioned this pull request Oct 12, 2016

Benchmark fixes #9064

Closed

2 tasks

evanlucas force-pushed the nexttickperf branch from fc514fd to 804d57d Compare October 13, 2016 17:17

evanlucas closed this Oct 13, 2016

evanlucas deleted the nexttickperf branch October 13, 2016 17:18

evanlucas added the lts-watch-v4.x label Oct 13, 2016

evanlucas merged commit 804d57d into nodejs:master Oct 13, 2016

MylesBorins added the lts-watch-v6.x label Nov 11, 2016

MylesBorins added dont-land-on-v4.x and removed lts-watch-v4.x labels Dec 21, 2016

MylesBorins mentioned this pull request Dec 21, 2016

V6.9.3 proposal #10394

Merged

evanlucas mentioned this pull request Dec 23, 2016

process: improve performance of nextTick (v4.x backport) #10433

Closed

2 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

process: improve performance of nextTick #8932

process: improve performance of nextTick #8932

evanlucas commented Oct 4, 2016

Fishrock123 commented Oct 4, 2016

evanlucas commented Oct 4, 2016

claudiorodriguez Oct 5, 2016

evanlucas Oct 5, 2016

Fishrock123 Oct 5, 2016

evanlucas Oct 5, 2016

Fishrock123 Oct 5, 2016

evanlucas Oct 5, 2016

AndreasMadsen commented Oct 5, 2016 •

edited

Loading

evanlucas commented Oct 5, 2016

claudiorodriguez commented Oct 5, 2016

claudiorodriguez left a comment

Fishrock123 commented Oct 5, 2016

evanlucas commented Oct 5, 2016

jbergstroem commented Oct 5, 2016

evanlucas commented Oct 5, 2016

AndreasMadsen commented Oct 5, 2016

imyller left a comment

imyller commented Oct 5, 2016

jbergstroem commented Oct 5, 2016

evanlucas commented Oct 5, 2016

lpinca left a comment

trevnorris commented Oct 6, 2016

jasnell left a comment

evanlucas commented Oct 10, 2016

evanlucas commented Oct 11, 2016

AndreasMadsen commented Oct 11, 2016

evanlucas commented Oct 13, 2016

MylesBorins commented Nov 11, 2016

evanlucas commented Nov 11, 2016

MylesBorins commented Dec 21, 2016

evanlucas commented Dec 23, 2016

process: improve performance of nextTick #8932

process: improve performance of nextTick #8932

Conversation

evanlucas commented Oct 4, 2016

Checklist

Affected core subsystem(s)

Description of change

Fishrock123 commented Oct 4, 2016

evanlucas commented Oct 4, 2016

claudiorodriguez Oct 5, 2016

Choose a reason for hiding this comment

evanlucas Oct 5, 2016

Choose a reason for hiding this comment

Fishrock123 Oct 5, 2016

Choose a reason for hiding this comment

evanlucas Oct 5, 2016

Choose a reason for hiding this comment

Fishrock123 Oct 5, 2016

Choose a reason for hiding this comment

evanlucas Oct 5, 2016

Choose a reason for hiding this comment

AndreasMadsen commented Oct 5, 2016 • edited Loading

evanlucas commented Oct 5, 2016

claudiorodriguez commented Oct 5, 2016

claudiorodriguez left a comment

Choose a reason for hiding this comment

Fishrock123 commented Oct 5, 2016

evanlucas commented Oct 5, 2016

jbergstroem commented Oct 5, 2016

evanlucas commented Oct 5, 2016

AndreasMadsen commented Oct 5, 2016

imyller left a comment

Choose a reason for hiding this comment

imyller commented Oct 5, 2016

jbergstroem commented Oct 5, 2016

evanlucas commented Oct 5, 2016

lpinca left a comment

Choose a reason for hiding this comment

trevnorris commented Oct 6, 2016

jasnell left a comment

Choose a reason for hiding this comment

evanlucas commented Oct 10, 2016

evanlucas commented Oct 11, 2016

AndreasMadsen commented Oct 11, 2016

evanlucas commented Oct 13, 2016

MylesBorins commented Nov 11, 2016

evanlucas commented Nov 11, 2016

MylesBorins commented Dec 21, 2016

evanlucas commented Dec 23, 2016

AndreasMadsen commented Oct 5, 2016 •

edited

Loading