Closed Bug 629278 Opened 14 years ago Closed 12 years ago

Typing into Twitter "Whats Happening" text box EXTREMELY slow (entry or delete). Typing maxes CPU.

Categories

(Core :: General, defect)

x86
All
defect
Not set
normal

Tracking

()

RESOLVED WORKSFORME
Tracking Status
blocking2.0 --- -

People

(Reporter: enger, Unassigned)

References

()

Details

(Keywords: qawanted)

User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US) AppleWebKit/534.10 (KHTML, like Gecko) Chrome/8.0.552.237 Safari/534.10 Build Identifier: Minefield 4.0b11pre (2011-01-26) Typing into the box to create a new tweet is VERY sluggish. Even using the "backspace" key to delete text in that box is VERY slow. I used the Task Manager and saw that if I hold down a key (trying to input text continuously (at the keyboard/system keystroke repetition-rate) into the box, or delete continuously from the box) that the CPU will be maxed-out! (even though visible repetition rate is VERY slow. This issue has something to do with how they engineer the text box on Twitter. I tried a few other web pages and there is no sluggishness input/deleting text. Twitter must be doing something cpu-intensive to manage their "Whats Happening" (compose a twitter message) text box. Whatever software or technique that Twitter is using for that text-box, Firefox and Minefield don't like it. I tried this experiment with IE8, Chrome 8.0.552.237, FF 3.6.13 and Minefield 4.0b11pre(2011-01-26). Hold down any key and let the letters appear into the "What's happening" box; watch the CPU column in the Windows Task Manager processes table. IE8 and Chrome appear to keep up with the key repetition rate of the computer. They are not sluggish. IE8 process uses 40%. Chrome uses 38% in one process, and 4% in another chrome process. FF 3.6.13 is SLOW. Cannot keep up with key repetition rate of computer. It consumes 50% of the cpu (and 4% additional is consumed by the system process). Minefield is BEYOND SLOW: glacial-pace. It's like 3 characters per second or worse. AND, it paints a character or two AFTER you have taken your finger off the key you were pressing (it is processing a backlog of keystrokes that it cannot keep up with!). It consumes 48 to 49% of CPU (and 6% to 8% on the system process too). Note: my dumb hardware is dual core (one thread per core). So, when task-manager reports 50% usage by a process, it is essentially telling us that the process is using as much CPU as it can get (assuming single-execution-flow, non-threaded process, etc) Minefield is thus pegging the system when I hold down a key, entering repeated keystrokes into the Twitter What's Happening text box. IE8 and Chrome don't peg the CPU and operate (apparently) at full speed (keeping up with the keyboard keystroke repetition-rate). Reproducible: Always Steps to Reproduce: 1. Start browser of choice on one monitor (I used full screen on external monitor) 2. Start Task Manager. Select Processes tab. click on CPU column twice so it sorts high-usage process to the top. (I used the main monitor on my laptop for this) 3. Go to twitter and log into your account 4. Click mouse into the "Whats Happening" text box (so you can pretend to compose a new tweet) 5. Push a regular letter-key on your keyboard and keep it pressed. 6. Note the rate at which characters are being painted into the text box 7. Note CPU attributed to Browser process(s) and system process 8. repeat experiment on Twitter page using different browsers, record results 9. Repeat similar test using text-boxes on other web sites. I didn't see any problems, with any of the browsers; BUT, it does allow you to observe the native keystroke repetition-rate of your keyboard/system. You can then compare this to your experiences using the What's Happening text box on the Twitter page. Actual Results: Firefox is slow. Minefield is MEGA SLOW. Keystrokes paint to screen slower than repetition-rate of keyboard/system. HIGH cpu usage while typing new text or deleting. In comparison, IE8 and Chrome seem to keep up with the keystoke repetition rate of my keyboard/system (and don't eat up an entire core's worth of cpu). Expected Results: Typing text or deleting text happens at the full keystroke repetition-rate of the keyboard/system, and without chewing up a huge amount of cpu. windows-xp SP3 32-bit with all the current patches. 4Gigs of RAM (3+ are recognized and used) No other applications running while doing the tests. Ancient crappo hardware: Dell Latitude D820 laptop w/ 1920x1200 internal display. Separate 1920x1200 external monitor connected to VGA port. NVIDIA Quadro NVS 120M graphics chip. Intel T2600 CPU at 2.16Ghz DirectX 9.0c I also tried restarting Minefield with extensions disabled. This didn't seem to make any difference.
Summary: "Whats Happening" text box EXTREMELY slow (entry or delete). Typing maxes CPU. → Typing into Twitter "Whats Happening" text box EXTREMELY slow (entry or delete). Typing maxes CPU.
Status: UNCONFIRMED → RESOLVED
Closed: 14 years ago
Resolution: --- → DUPLICATE
Robert says that this is not fixed for him, even though bug 579488 is fixed. So something else is going on here.... Robert, does disabling hardware acceleration make a difference for you?
Keywords: qawanted
Status: RESOLVED → REOPENED
Ever confirmed: true
Resolution: DUPLICATE → ---
The 2011-01-27 nightly contained an important fix here, and since this bug was filed on 2011-01-26 I'd like to ask the reporter to retest to see if anything changed.
He did, with a 2011-01-29 build. See bug 579488 comment 64.
Sure, he mentioned the cpu usage issue, but he didn't mention how it feels to a user like he did above. Is it just as slow as before or not?
blocking2.0: --- → ?
Product: Firefox → Core
QA Contact: general → general
I'd also be interested in how a current nightly compares to 3.6 in that regard.
hi: Sorry for the delay in responding. I hope this information adds some precision to the report. Minefield updated again. It is now: 4.0b11pre (2011-01-30). Google Chrome is 8.0.552.237 IE8 is version 8.0.6001.18702 FireFox is Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9.2.13) Gecko/20101203 Firefox/3.6.13 ( .NET CLR 3.5.30729; .NET4.0E) I thought some numerical precision might help. So, I got out a stop watch and ran some tests. I basically held down a key for 10-seconds and then counted how many characters appeared on the screen. Dividing by 10 yields the characters per second (CPS) rate. I repeated each test TWICE. Rather than tracking the CPU usage per process, I simply report the amount of idle remaining. Recall this is a dual-core, one thread-per-core system. The system normally sits reliably at 99% idle. Reporting 50% idle means that an entire core (or equivalent across the two cores) is fully consumed. (for a single, non-threaded activity, that's as much as you can get) I tested by typing into CommandWindow, NotePad and the "what's Happening" box in Twitter. I use CommandWindow and NotePad to baseline the native keystroke repetition rate of the system. The test of the "What's Happening" text box then shows how much slower it is IN REAL NUMBERS (not just gut feeling) AND it shows how much burden is being placed on the computer by the TWITTER text entry software. CommandLine 31 and 32 CPS 99% idle both times NotePad 32 and 33 CPS 99% idle both times MineField 16.5 and 16.5 CPS 50% idle both times GoogChrome 32.6 and 32.5 CPS 49% idle both times IE8 21.4, 22.6, 22.2, 20.5 varying 49 or 50% idle each time FireFox 11.3, 11.0, 11.1, 11.0 50% idle each time (I did IE8 and FF tests four times, because the CPS values didn't cluster as tightly.) On a lark, for Minefield only, I also tried using BACKSPACE in the "what's Happening" box (to remove the characters I put there on the second trial). In 10 seconds it removed all but one character. So the BackSpace rate is also approx 16.5 cps. And, the CPU was at 50% idle. In summary: The native keystroke repetition rate is approximately 32CPS. Using native windows apps, sustaining that input rate consumes NO CPU. When inputting characters into the twitter "what's happening" text box: All the browsers consume a FULL CORE. (Chrome may consume 1% of the 2nd core?) Chrome keeps up with the full 32 CPS character repetition rate. No other can. IE8 can do about 2/3 of the rate Minefield can do 1/2 of the rate Firefox can do 1/3 of the rate. I hope these numbers are more useful than a vague report that "x or y seems faster". Minefield is concretely faster than FF. But, they don't even run as fast as dumb old IE8, much less Google Chrome. (What in the world are the Twitter people thinking? An entire CPU core consumed just to input text?!?! )
Boris: Regarding hardware acceleration, it appears to already be turned OFF. Here is the GRAPHICS section from the bottom of the "about:support" listing of Minefield: Graphics Adapter DescriptionNVIDIA Quadro NVS 120M Vendor ID10de Device ID01d8 Adapter RAMUnknown Adapter Driversnv4_disp Driver Version6.14.11.5683 Driver Date11-16-2007 Direct2D EnabledBlocked on your graphics driver. Try updating your graphics driver to version 257.21 or newer. DirectWrite Enabledfalse (0.0.0.0) WebGL Renderer(WebGL unavailable) GPU Accelerated Windows0/1
Robert: thank you for that thorough experimentation; it's very helpful. I suspect that IE8 is faster because it simply doesn't implement the shadow and other effects that cause significant overhead, but I could be wrong about the particulars. Could you try updating your video drivers, and see if that helps? Would be good to know what impact (positive or negative) D2D acceleration has.
(In reply to comment #7) > I thought some numerical precision might help. So, I got out a stop watch and > ran some tests. I basically held down a key for 10-seconds and then counted > how many characters appeared on the screen. Dividing by 10 yields the > characters per second (CPS) rate. I repeated each test TWICE. Yes! This is fantastic data! Thanks for taking the time to test. What I was glad to learn from your data is that Minefield _is_ faster than Firefox 3.6. This is good, but we are not fast enough yet. I knew that we needed to do better on Twitter, and we have plans to make it a lot faster in the future (display list based invalidation, making box-shadow much faster are the biggest ones), but not in time for FF4.
Mike: I visited the Nvidia site last night. Their support site does list the video module used in my laptop. But, they do NOT list any driver for it. (nothing at all: new, old, beta or WindowsValidated). The 120m is one of those hacks they build for OEM integrators, not an independently released product. What I need to do is ****-can the entire laptop. It is a fugitive from the Smithsonian as it is. BUT, it is worth noting that many regular users (non early adopters, non computer wizards) are stuck using this older generation hardware. Both my wife and I have HP laptops from the IT departments at our two companies that are so old that they independently receive AARP invitations. Good or bad, Firefox should probably be able to support this old junk, at least for a while longer. Consuming an entire CPU core while typing is beyond the pale. What happens as twitter evolves. Will we need dual GTX580s in SLI to load their web page? :-)
OS: Windows XP → All
Thanks Robert. Given we've improved since Firefox 3.6, we shouldn't block FF4 on making it even better.
blocking2.0: ? → -
Twitter made a change to their site recently that should make this a lot faster.
Assignee: nobody → tnikkel
Assignee: tnikkel → nobody
Mark the bug as resolved?
Status: REOPENED → RESOLVED
Closed: 14 years ago12 years ago
Resolution: --- → WORKSFORME
Whilst this works okay for standard browser text boxes, it is still about as bad as ever on some applications. Best example is the game War Commander on Facebook. Easy to check and replicate, so I really don't know why an attempt to investigate in this manner has not as yet been attempted.
Because this bug was filed on Twitter. If typing into some other textbox is slow, that should be filed as a different bug.
You need to log in before you can comment on or make changes to this bug.