Skip to content
This repository has been archived by the owner on Apr 22, 2023. It is now read-only.

Node 0.8.9 to 0.10.21 cpu impact. #6447

Closed
vsm83 opened this issue Nov 1, 2013 · 3 comments
Closed

Node 0.8.9 to 0.10.21 cpu impact. #6447

vsm83 opened this issue Nov 1, 2013 · 3 comments

Comments

@vsm83
Copy link

vsm83 commented Nov 1, 2013

I've got one-thread server on amazon aws (x-large) which use "net" TCP server, having about 2700 open active sockets in peak time.
When sockets count raising to 2800, cpu load is 100% and database connections (pg module) begin lagging.
But when I`ve changed node version down to 0.8.9, CPU usage decreased to 20% (with periodic 100% GC peaks)
What's happened with node 0.10.21?
All config options are default.

@bnoordhuis
Copy link
Member

Hard to say unless you have a test case or example app that uses core modules only (which doesn't seem to be the case as you mention the pg module.) Please use the mailing list for more support.

@vsm83
Copy link
Author

vsm83 commented Nov 1, 2013

I used profiler "nodetime".
With profiler I`ve found that in v.0.10.21 the most of CPU used for exports._unrefActive in timer.js, but in v. 0.8.9 there is no exports._unrefActive
111
222
333

@bnoordhuis
Copy link
Member

_unrefActive() was added in v0.10.10 in commit f46ad01 and is a known bottleneck. It's scheduled for optimization in v0.12. Commit joyent/libuv@21c37a7 from earlier this week is one such optimization.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants