搜尋 Mozilla 技術支援網站

防止技術支援詐騙。我們絕對不會要求您撥打電話或發送簡訊,或是提供個人資訊。請用「回報濫用」功能回報可疑的行為。

Learn More

Performance Regression: After update to FireFox 18 permanent high CPU load

  • 38 回覆
  • 68 有這個問題
  • 4 次檢視
  • 最近回覆由 John99

more options

After the last update FireFox started to continuously eat a whole CPU on my 2-core laptop. My usage pattern (regarding browsed sites and/or activated plugins) has not changed, but I must confess that I'm a heavy user of plugins.

I suspect the problem may arise from a changed performance behaviour of a plugin due to the new JS engine of FF18, but how can I easily trace this problem further down. I.e. is there a way to inspect the collected performance data of firefox (activated)?

After the last update FireFox started to continuously eat a whole CPU on my 2-core laptop. My usage pattern (regarding browsed sites and/or activated plugins) has not changed, but I must confess that I'm a heavy user of plugins. I suspect the problem may arise from a changed performance behaviour of a plugin due to the new JS engine of FF18, but how can I easily trace this problem further down. I.e. is there a way to inspect the collected performance data of firefox (activated)?

被選擇的解決方法

The upgrade from FF18 to FF19 seems to have solved the problem for now. Performance is much better, even with many open tabs. CPU usage does peak over 90% from time to time, but doesn't stay there longer than a few seconds.

However, I still have the about:config javascript.options.methodjit preferences set to FALSE. I'm going to leave them that way for the time being, until I'm satisfied that the upgrade has made the performance consistently better than it was with FF18. Once I'm satisfied with that, then I'll reset those preferences to TRUE.

I'll post here again if performance goes back down the dumper, with or without those preferences set to TRUE.

Thanks to all who took the time to read my posts and respond!

Alan Mintaka

從原來的回覆中察看解決方案 👍 0

所有回覆 (18)

more options

If about:telemetry is available then you should see it listed on the about:about page (it is there in Firefox 19).

more options

@cor-el,

I'm not sure if you're answering my last post. If you are, I was evidently still editing my message when you posted. I did manage to get about:telemetry to work by installing the add-on for FireFox versions older than 19 (mine is 18).

I'm going to leave 18 installed until I can figure out what to do with the telemetry data. However - would upgrading to 19 blow away the telemetry for 18? I would like to keep it if I can, in the event it proves useful somehow.

Also, note some of the messages I listed in my previous post. Regardless of what all of the barcharts mean, slow SQL response was flagged on the "Main" and "Other" threads.

Thanks for your time,

Alan Mintaka

由 AlanMintaka 於 修改

more options

I am not sure if any of this will help at all, but it may at least point you at some information sources.

Sorry I do not really understand what all the data is myself, although I can see there is an awful lot of information. All I know about where it is stored is that I can see files in my firefox profile that apparently relate to the telemetry add-on and to the data itself, such as a *savedTelemetry.dat.json apparently includes the raw data for the histograms.

I am not sure if saving old versions of that file and overwriting (rename files as necessary maybe) the current file with it could be used as a method of preserving and viewing the info.

My initial thought was that merely comparing your data with the pooled data probably some would show up as very bad and would at least be quntitative data you would be able to quote when expressing how bad performance was.

I am not a developer but I do note developer info shows it is possible to design specific histograms, and as a separate subject there is the profiler that may be used ( ? on custom builds ? )

  1. https://developer.mozilla.org/en-US/docs/Performance/Adding_a_new_Telemetry_probe
  2. https://developer.mozilla.org/en-US/docs/Performance/Profiling_with_the_Built-in_Profiler
    & https://developer.mozilla.org/en-US/docs/Performance/Profiling_with_the_Built-in_Profiler_and_Local_Symbols_on_Windows

I am not sure where this may be discussed, although there are a wealth of Mozilla forums and mailing lists.

more options

Thanks john99,

Those links look pretty good, especially the one for "adding_a_new_telemetry_probe". I didn't know such things could even be done.

For now I'll concentrate on the parts of the telemetry I can read, such as "Slow SQL Response".

Also, and as usual, I wasn't thinking. I'll just save the current about:telemetry page for FF 18 on my HDD and then upgrade to FF19. Who knows, the upgrade may solve the problem on this client.

Thanks for the info,

Alan Mintaka

more options

選擇的解決方法

The upgrade from FF18 to FF19 seems to have solved the problem for now. Performance is much better, even with many open tabs. CPU usage does peak over 90% from time to time, but doesn't stay there longer than a few seconds.

However, I still have the about:config javascript.options.methodjit preferences set to FALSE. I'm going to leave them that way for the time being, until I'm satisfied that the upgrade has made the performance consistently better than it was with FF18. Once I'm satisfied with that, then I'll reset those preferences to TRUE.

I'll post here again if performance goes back down the dumper, with or without those preferences set to TRUE.

Thanks to all who took the time to read my posts and respond!

Alan Mintaka

由 AlanMintaka 於 修改

more options

Glad to see there is some possible improvement.

Might be interesting to see if it is possible to work out which bug may have remedied this, I do not see anything obvious in

On the subject of bugs I see the one filed earlier

  • Bug 832864 - FF18 uses much more CPU load than FF17 depending on the number of opened tabs

and mentioned in this thread was closed as not actionable due to lack of: Info & valid STR.

more options

Another tool for checking performance is the new profiler. It is still in development but is being integrated into Firefox and so far has made it into Firefox 20 on the Beta channel, so may be only six weeks away.

more options

Original Poster here: I'm still monitoring the behaviour, but similar to Alan it seems, that the latest update to FireFox 19.x may have resolved the issue. If not - I will report back hopefully armed with more details.

As a software developer I'm really keen & looking forward to the upcoming built-in FireFox profiler.

I really want to thank all the people here for the really helpful and in-depth support given here! Even if we could no clearly pin down the issue, this question provided a lot of very helpful pointers and insights into FireFox and their roadmap. My experience outstands all my previous support experiences I ever had, even with highly-paid commercial solution providers.

Many kudos here especially to John99 and AlanMintaka for their extensive research.

more options

You are welcome, pity we did not get to the bottom of the issue, but most importantly it seems to be resolved by the upgrade.

more options

I'm sorry, but I have to disagree. The issue isn't really fixed.

Yes, FF19 uses less CPU load than FF18, but still much more than FF17 did.

Here my average values: The System: i5-CPU, 16GB RAM, Win7 x64

FF17.01: CPU-load 1.0% average at 300 tabs

FF18.02: CPU-load 10.0% average at 300 tabs

FF19: CPU-load 4.5% average at 300 tabs

So FF19 uses still 3 times more CPU load than FF17 did, both in normal mode and in safety mode with disabled addons. I my opinion, that's not enough to mark the problem as solved.

由 1394 於 修改

more options

@1394, What about time slices though? FF19 may be using that much CPU, but does it give up time slices for other processes more efficiently than FF18?

Here's how I opened my last post: "The upgrade from FF18 to FF19 seems to have solved the problem for now. Performance is much better, even with many open tabs. CPU usage does peak over 90% from time to time, but doesn't stay there longer than a few seconds."

My original complaint about the high CPU usage was in reference to how it slowed system performance overall and crippled the other applications that were running. That's not happening with FF19, at least not in any way that I can measure. All I know is that the overall system performance problems went away when I updated to FF19, even though FF19 occasionally peaks the CPU over 90%. That's good enough for me.

more options

@AlanMintaka, well, it's good, that you are now able to work again with your computer with opened Firefox.

For average CPU usage I don't measure the peaks, which still occur from time to time. The peaks can go up to full core usage, so in my case with a 4Core CPU up to 25%.

Once again, good, that with the new upgrade many people can use Firefox furthermore, but the core of the problem doesn't seem to be solved yet. FF19 is still using much more CPU load than FF17, even you are now able to use your PC again.

more options

1394
Have you tried using a clean profile and running in safe mode with no plugins enabled ?

I you are prepared to provide evidence of your figures, and run tests if required then you could consider politely posting your assertion about the performance regression as a comment in one of these blogs

You may get a reply from someone with useful advice.

more options

@John99

Jep, new profile, fresh installation, safe mode, disabled plugins, disabled addons - I've tried everything. FF17 is working well, FF18 eats CPU and FF19 is inbetween.

Are you able to review your average CPU usage with FF19 while opened about 300 tabs? Doesn't matter which ones, just open tabs and check, if your average CPU load gets high. I really think, that the most people are satisfied when they can use their PC in normal way. So they don't care about a few percentage CPU load more or less as long as the PC is still running. However, I'll stay at FF17, but I think I'll change to ESR as long as there is no solution in sight.

Btw, many thanks for your hints, but I'm tired in spending time to try getting FF working in the same way as it's did before they rebuilt the whole JS engine and maybe much more.

more options

I just tried fx20 with 500tabs in one window and another hundred or so other tabs in other windows. Each of the cores stayed under 25% . That could be considered high but hardly causing problems.

Try the blog site I mentioned, you could have come across some as yet unknown regression.

more options

Well, many thanks for testing your FF. So the issue, if Mozilla would call it so, is still alive in FF20. Of course, this issue dosn't cause big problems, you can work furthermore, even, if one of four cores stays permanently on 25% percent. But FF isn't the light-cpu-using-browser anymore as it was before.

Ok, I'll try your blog-site ;) Thanks again

more options

@1394 Opposed to the issue I and Alan had where we actually failed to provide a solid bug report, your issue seems to be very easily reproducible. Please consider that FireFox is a free project and its relevance by user base is vital for keeping the Internet free! Therefore I'd join John99 kind encouragement, if you as a stakeholder in the tab issue could try to bring up this into attention at the right places. This would be great!

more options

1394

 before they rebuilt the whole JS engine 

Note the JIT compiler may be disabled by toggling a pref as mentioned upthread, or by using Safe Mode, so your own tests already suggest this is unlikely to be the cause of any CPU load increase you see.

  1. 1
  2. 2