-
|
ran some benchmarks with nvm and it does appear there is correlation with each new node version and decreasing performance of uWS.js http (requests per second)
|
Beta Was this translation helpful? Give feedback.
Replies: 14 comments 14 replies
-
|
Absolutely. That's what I've been saying for years. The only limit uWS.js has is They never benchmark their changes and when they do, they still allow small drops in perf. by every release, because they have no discipline or leadership to refuse offending commits. Many drops make an ocean over time. |
Beta Was this translation helpful? Give feedback.
-
|
Websocket echos per second also showing the same thing. What is going on over there they just keep adding more code and it is slowing stuff down?
|
Beta Was this translation helpful? Give feedback.
-
|
Must be something they can do to fix it, that's a 25% performance drop for sockets and 20% drop for http from node v10 and this library only hooks into theirs minimally so maybe have someone over there look into it, get it back up to 100% |
Beta Was this translation helpful? Give feedback.
-
|
If you look at the history of (C++) uWS you see that the trend went the other way: v0.14.8: v18.9.0: This is for WebSockets but the same can be seen in Http, even though v18 has a complete router on top. We've remained at a flat line since the jump from v0.14.8 -> v0.15.0, because we do not allow small drops to slowly form an ocean over time. If someone sends a PR that decrements perf. I refuse it. |
Beta Was this translation helpful? Give feedback.
-
|
I wrote a benchmark that just does
And this benchmark does nothing at all besides calling If you then look at the vanilla (Google) V8 counterpart we get 104ms in Node.js 8, and 136ms in Node.js 14.5.0. So anything above 200ms is definitely a regression caused by Node.js itself, and not V8 or any native addon. |
Beta Was this translation helpful? Give feedback.
-
|
@BridgeAR You had some questions regrading this a while back - what's your take on this? |
Beta Was this translation helpful? Give feedback.
-
|
For the visual learners |
Beta Was this translation helpful? Give feedback.
-
|
Alright, nothing we can do here. |
Beta Was this translation helpful? Give feedback.
-
|
Apparently @trevnorris stopped working on nodejs core 3 years ago, and I guess @addaleax took over responsibility for those parts. This maps well with the graph; a slow but sure decline in perf. with every release |
Beta Was this translation helpful? Give feedback.
-
|
I did manage to reach Trevor Norris on this and he's made a very detail per-commit benchmark and found the most offensive commit in this. That's great but unfortunately Trevor is not working on Node.js anymore, so I don't know if people will listen to him on this. https://gist.github.com/trevnorris/35c8e570c3097f58739b84bfa3f0a390 |
Beta Was this translation helpful? Give feedback.
-
|
This is the most offending commit: nodejs/node#30959 It's ironic that they claim to have run benchmarks when they clearly did not! |
Beta Was this translation helpful? Give feedback.
-
|
Trevor stats for
|
Beta Was this translation helpful? Give feedback.
-
|
@hst-m Is this still a problem in node v17? If that's still the case then why shouldn't we create an issue on nodejs repo? |
Beta Was this translation helpful? Give feedback.
-
|
@hst-m any chance of getting some new benchmarks with v15 through v17 please? |
Beta Was this translation helpful? Give feedback.
-
|
@lewispham Feel free to create an issue on the node repo. Explain the concern and give some benchmark numbers. I suspect some of the overhead is due to async hooks. Maybe ask if there is a possibility to disable this for those that don't care about async context and just want maximum performance. I believe Alex has created issues for this in the past so I don't know if it's likely it to be changed. But it was quite some time ago and I don't think it would hurt to have another go at discussing it. I tried taking a look at it today but it's unfortunately out of my depth. |
Beta Was this translation helpful? Give feedback.
-
|
Yes, I have noticed some communication challenges. I'll have go at your patch and see what I can do. Thanks. |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
For information, last nodejs version https://github.com/nodejs/node/releases/tag/v17.4.0 includes the perf improvement implemented in nodejs/node#41331 which I think adress that issue. |
Beta Was this translation helpful? Give feedback.
-
|
I ran benchmarks back to Node 14, I see 18% improvement since this issue was created, and the commit at 17.4 helped. The lowest point was Node 14.5, and the latest Node 17.4 is the fastest since Node 12. It's about half way back to the Node 10 high. So it's good to see a positive direction
|
Beta Was this translation helpful? Give feedback.
-
|
Yes just following up for these people
I was half expecting to see performance continue a nosedive in the downward direction, but was pleasantly surprised to see it reverse course |
Beta Was this translation helpful? Give feedback.
-
Personnaly I'm a bit disappointed by these figures, I was expecting from nodejs/node#41331 a greater improvement than 3% |
Beta Was this translation helpful? Give feedback.
-
|
You can't polish a turd |
Beta Was this translation helpful? Give feedback.
-
|
The main problem in Nodejs concerning performance issues like this one is that there no automatic performance regression checks, the benchmark WG has been decommissionned, and there's only one core contributor, mscdex, who is actively tracking performances issues/regressions. |
Beta Was this translation helpful? Give feedback.
-
|
It is what it is |
Beta Was this translation helpful? Give feedback.
Absolutely. That's what I've been saying for years. The only limit uWS.js has is
node::MakeCallbackand they manage to tamper with that every release, to our disadvantage.They never benchmark their changes and when they do, they still allow small drops in perf. by every release, because they have no discipline or leadership to refuse offending commits. Many drops make an ocean over time.