MozillaZine


Possibly weird "fuzzy" topic on resource utilization

Discussion about Seamonkey builds
RDaneel
 
Posts: 532
Joined: January 19th, 2004, 2:43 pm
Location: Puget Sound, WA

Post Posted March 19th, 2021, 2:52 pm

Nope, this isn't about the amount of memory used by SeaMonkey. ;)

Months ago (6? more?) I noticed something that I found personally troubling on one of the nightly builds: the "context switches" per second on the SM process went up dramatically (like from single digits to 200-300). :(

I am bringing this up now because with yesterday's nightly 2.53.8b1pre [64-bit] build, the numbers went back down to single digits. :)

For a bit more context, we are talking my usual ~200+ tabs environment - but as long as you don't "activate" a tab by forcing it to become current, it appears to keep the resource usage down a lot. I derive my numbers from my two favorite go-to system utilities: Process Hacker and HWiNFO64.

Remember, I said in the subject that this was going to be "fuzzy", as in no hard numbers and cited studies... but, why do we care?

Modern CPUs (and the OSes tuned to work with them) are all about "bursty" workloads, and this matters for way more than just the "mobile" cases. The CPUs like to do things like shutting down entire sections of logic, and drop the voltages on the portions that are still left "active" - all on fairly short time-scales, I believe on the order of milliseconds.

Still, why does this matter? Well, aside from concerns of "engineering elegance", this is what allows for the so-called "turbo"/"boost" modes of operation... when coupled with the readings from numerous internal temperature sensors, the CPUs are able to "pump up" their clocks (and subsequent power usages) from, say, 800 MHz to 4 GHz+, in response to momentary demand, and get some work done - and then bring everything back down to the pseudo-napping state.

This "cool" model (pun intended) is broken by things like polling behavior, since the CPU is not able to drop as far down into (or spend as much time in) various sleep states, which means the power utilization and connected temperatures both stay up - blocking the boosting of clocks and voltages.

Since I don't know which checkins are responsible for the ugly jump to hundreds of "context switches" - or which ones brought SM back to the more elegant single-digit levels - I can only report that according to the internal power usage readouts from my Core i7 "Kaby Lake" processor, we were never able to drop below ~800+ mW for months, but now can hover around ~500+ mW. Put another way, even a "quiescent" SM was the usual big user of CPU, and now it doesn't make the top 5.

So, thanks! And if it is clear which changes caused this / fixed things, I really hope they can stay around. ;)

frg
 
Posts: 1149
Joined: December 15th, 2015, 1:20 pm

Post Posted March 19th, 2021, 3:09 pm

Might actually https://bugzilla.mozilla.org/show_bug.cgi?id=1699322 too or https://bugzilla.mozilla.org/show_bug.cgi?id=1698675.

Seems the render code thought some of the gifs were one frame animated gifs.

FRG

ndebord

User avatar
 
Posts: 1052
Joined: December 7th, 2002, 9:53 am

Post Posted March 20th, 2021, 5:12 pm

frg wrote:Might actually https://bugzilla.mozilla.org/show_bug.cgi?id=1699322 too or https://bugzilla.mozilla.org/show_bug.cgi?id=1698675.

Seems the render code thought some of the gifs were one frame animated gifs.

FRG


frg,

So, read the bug... saw that only one cz.gif in the xpi file. Changed it to png. Is that a problem in 2.53.6 or should I wait until 2.53.8?

Nick
Last edited by ndebord on March 21st, 2021, 8:35 am, edited 1 time in total.
-N- Quis custodiet ipsos custodes
FrameWork, SeaMonkey(32bit),Windows 10 Pro (X64- 21H1), WinPatrol, Malwarebytes & AVG Suite

frg
 
Posts: 1149
Joined: December 15th, 2015, 1:20 pm

Post Posted March 21st, 2021, 5:33 am

cZ Bug 1698675 will end up in 2.53.7 final probably today or tomorrow. I would just pick 2.53.7b1 today and install the final over it. No regressions reported and other than this bug only 2 build fixes which might help FreeBSD building will ned up in the final.

ndebord

User avatar
 
Posts: 1052
Joined: December 7th, 2002, 9:53 am

Post Posted March 21st, 2021, 8:34 am

frg wrote:cZ Bug 1698675 will end up in 2.53.7 final probably today or tomorrow. I would just pick 2.53.7b1 today and install the final over it. No regressions reported and other than this bug only 2 build fixes which might help FreeBSD building will ned up in the final.


frg,

Thanks, guess it is time to update to 2.53.7xxx

Appreciate,

Nick
-N- Quis custodiet ipsos custodes
FrameWork, SeaMonkey(32bit),Windows 10 Pro (X64- 21H1), WinPatrol, Malwarebytes & AVG Suite

Return to SeaMonkey Builds


Who is online

Users browsing this forum: No registered users and 1 guest