Still a significant CPU ratcheting effect when changing PublisherIDs

[tracedwax]

Hi Arun, Nigel, & co,

Thanks again for making the bug fixes last week -- seems to have improved the ratcheting issue some. 

Running a battery of further tests this morning, I noticed however that the improvement wasn't as much as I'd originally hoped: there was still a large ratcheting effect with a significant impact.

To isolate the problem, I ran your swf again at http://blogs.adobe.com/arunpon/files/2011/05/WebCameraFinalTrace2.swf.  My standard test has been to start the app on 4 computers, turn on cameras for the first 3 computers, and take an initial reading of CPU usage.  Then I measured the CPU each time I clicked on someone to put them on the larger WebcamSubscriber, then switched once per second between them, in order, ten times each.  Then I turned the fourth computer's camera on and measured its CPU usage.  I then waited 10 minutes and measured everyone's CPU usage, then switched 10 more times over an additional 30 seconds.

We ran experiments like this multiple times and found noticed significant CPU ratcheting.  CPU usage would recover sometimes, but the recovery took a long time.  The fourth computer showed no CPU increase until its camera turned on, at which point it showed a very large CPU increase.  Putting the fourth computer on the large WebcamSubscriber crashed Flash Player.  Waiting for 10 minutes, the CPU hadn't decreased for any of the computers.  Repeating the switching, however, the computers' CPU initially decreased by a lot (perhaps garbage collection kicked in), but then increased again as before.

Here are the details.  Computer A is a MacBookPro '10 i5 4GB, Computer B is a MacBook '09 CoreDuo 2GB, Computer C is a MacBook Pro '10 i7 8GB, and Computer D is a '07 MacBook Pro CoreDuo 3GB.  They're all Macs, but PCs showed similar ratcheting effects.  A, B, and C were running Firefox 4 and D was running Firefox 3, but we've seen similar effects in other browsers.

The numbers below are the CPU percentages of the Flash Player process while running your SWF above that includes your publisherIDs bug fix:

Before starting cams

A - 1.6%

B - 2.1%

C - 2.5%

D - 6.5%

A,B,C start their cams

A - 10%

B - 10%

C - 9%

D - 22%

A is published to big screen

A - 19%

B - 12.5%

C - 12%

D - 33%

B is published to big screen

A - 13.5%

B - 22%

C - 12%

D - 32%

C is published to big screen

A - 14%

B - 13%

C - 19%

D - 32%

Switch between A,B,C 9 times each, such that A is on big screen

A - 36%

B - 17%

C - 24%

D - 33%

Switch to B on the big screen

A - 19%

B - 36%

C - 25%

D - 32%

Switch to C on the big screen

A - 20%

B - 19%

C - 63%

D - 33%

D turns on Cam

A - 26%

B - 24%

C - 64%

D - 113%

D to big screen

A - 26%

B - 24%

C - 37%

D - 125%

Wait 10 minutes, D freezes, put A on big Screen

A - 28%

B - 17%

C - 20%

put C on big screen

A - 18%

B - 17%

C - 48%

Switch 40 times in 30 seconds, put A on big screen

A - 85%

B - 37%

C - 14%

Put B on big screen

A - 38%

B - 92%

C - 14%

put C on big screen

A - 39%

B - 40%

C - 22%

Thanks very much,

-Trace

[tracedwax]

Many thanks again, Nigel -- In our initial testing, this seems to have resolved our ratcheting issue!

Would love to also get the 10.3 source with AEC, we'll use that on Monday for our scheduled user trial / load test!

Thanks,

-Trace

[BARRYC16]

Hi Nigel,

I'll wait until there's a 10.3 swc seeing as it seems like the problem is fixed.

Thanks,

Barry

[aponnusa]

Ah Nigel. You beat me here, I wanted to fix this issue before you could.

I am attaching the 10_3 swc (both regular & FlashOnly) with AEC. It doesnt have the peer-assisted networking components.

RIP- CPU Ratceting for LCCS WebCam.

Thanks

Arun

[BARRYC16]

Thanks Arun,

I'll run a full test tomorrow.

Thanks,

Barry

[BARRYC16]

Hi Nigel, Arun,

I just ran a test on my app and we still see significant problems. I see the Flash player on one of my machines going up to 70% CPU usage. It also seems like the memory footprint keeps climbing. On all machines the Flash player seems to just slow down a lot.

This new version has definitely made a difference, but I think there are still issues here.

I can go back to the 10.0 swc to see if that fixes things or is that a pointless experiment?

Thanks,

Barry

[davisballen]

Hi Nigel,

This is Davis Allen - I'm also helping out on the youare.tv project.

Trace and I just ran some more tests with the new 10.3 beta swc and did indeed see a ratcheting effect. Both the initial cpu usage and the ratcheting were greater than on the 10.3 aec swc from the 10.5 lccs release.

We re-tested Nigel's 10.0 swc and again found that there was no ratcheting using that swc. Just to double check, did the 10.3 swc you provided contain the same fix as the 10.0 swc? If you repost the revised 10.3 aec beta swc we can re-test it immediately.

We really appreciate all the help,

Thanks,

Davis

[BARRYC16]

Hi Nigel,

Just to add to this. My tests were also done with the 10.3 swc. Didn't test with 10.0.

Thanks,

Barry

[aponnusa]

Hi,

Sorry all for the lost time. I forgot to integrate all the changes for the fix. I have built a new 10.3 swc. (I double verified it this time).

Thanks

Arun

[tracedwax]

Thanks, Arun.  Using this swc, our app doesn't ratchet.

So glad we got to the bottom of it -- very much appreciated by us and our users!

At your convenience, could you please also provide the source code corresponding to this 10.3 AEC swc?

Kind regards,

-Trace

[BARRYC16]

Thanks Nigel, Arun,

Looks like the problem is fixed for us also. Just ran the same test and CPU is fine.

Very much appreciated here also,

Barry