Bonum Certa Men Certa

Techrights Coding Projects: Making the Web Light Again

A boatload of bytes that serve no purpose at all (99% of all the traffic sent from some Web sites)

Very bloated boat



Summary: Ongoing technical projects that improve access to information and better organise credible information preceded by a depressing overview regarding the health of the Web (it's unbelievably bloated)

OVER the past few months (since spring) we've been working hard on coding automation and improving the back end in various ways. More than 100 hours were spent on this and it puts us in a better position to grow in the long run and also improve uptime. Last year we left behind most US/USPTO coverage to better focus on the European Patent Office (EPO) and GNU/Linux -- a subject neglected here for nearly half a decade (more so after we had begun coverage of EPO scandals).



As readers may have noticed, in recent months we were able to produce more daily links (and more per day). About a month ago we reduced the volume of political coverage in these links. Journalism is waning and the quality of reporting -- not to mention sites -- is rapidly declining.

"As readers may have noticed, in recent months we were able to produce more daily links (and more per day)."To quote one of our guys, "looking at the insides of today's web sites has been one of the most depressing things I have experienced in recent decades. I underestimated the cruft in an earlier message. Probably 95% of the bytes transmitted between client and server have nothing to do with content. That's a truly rotten infrastructure upon which society is tottering."

We typically gather and curate news using RSS feed readers. These keep sites light and tidy. They help us survey the news without wrestling with clickbait, ads, and spam. It's the only way to keep up with quality while leaving out cruft and FUD (and Microsoft's googlebombing). A huge amount of effort goes into this and it takes a lot of time. It's all done manually.

"We typically gather and curate news using RSS feed readers. These keep sites light and tidy. They help us survey the news without wrestling with clickbait, ads, and spam.""I've been letting wget below run while I am mostly outside painting part of the house," said that guy, having chosen to survey/assess the above-stated problem. "It turns out that the idea that 95% of what web severs send is crap was too optimistic. I spidered the latest URL from each one of the unique sites sent in the links from January through July and measured the raw size for the individual pages and their prerequisites. Each article, including any duds and 404 messages, averaged 42 objects [3] per article. The median, however, was 22 objects. Many had hundreds of objects, not counting cookies or scripts that call in scripts.

"I measured disk space for each article, then I ran lynx over the same URLs to get the approximate size of the content. If one counts everything as content then the lynx output is on average 1% the size of the raw material. If I estimate that only 75% or 50% of the text rendered is actual content then that number obviously goes down proportionally.

"I suppose that means that 99% of the electricity used to push those bits around is wasted as well. By extension, it could also mean that 99% of the greenhouse gases produced by that electricity is produced for no reason.

"The results are not scientifically sound but satisfy my curiosity on the topic, for now.

"Eliminating the dud URLs will produce a much higher object count.

“The results are not scientifically sound but satisfy my curiosity on the topic, for now.”
      --Anonymous
"Using more mainstream sites and fewer tech blogs will drive up the article sizes greatly.

"The work is not peer reviewed or even properly planned. I just tried some spur of the minute checks on article sizes in the first way I could think of," said the guy. We covered this subject before in relation to JavaScript bloat and sites' simplicity, but here we have actual numbers to present.

"The numbers depend on the quality of the data," the guy added, "that is to say the selection of links and the culling the results of 404's, paywall messages, and cookie warnings and so on.

"As mentioned I just took the latest link from each of the sites I have bookmarked this year. That skews it towards lean tech blogs. Though some publishers which should know very much better are real pigs:




$ wget --continue --page-requisites --timeout=30 --directory-prefix=./test.a/ https://www.technologyreview.com/s/614079/what-is-geoengineering-and-why-should-you-care-climate-change-harvard/ . . .

$ lynx --dump https://www.technologyreview.com/s/614079/what-is-geoengineering-and-why-should-you-care-climate-change-harvard/ > test.b

$ du -bs ./test.? 2485779 ./test.a 35109 ./test.b



"Trimming some of the lines of cruft from the text version for that article, I get close to two orders of magnitude difference between the original edition versus the trimmed text edition:

$ du -bs ./test.?
2485779	./test.a
35109	./test.b
27147	./test.c


"Also the trimmed text edition is close to 75% the size of the automated text edition. So, at least for that article, the guess of 75% content may be about right. However, given the quick and dirty approach, of this survey, not much can be said conclusively except 1) there is a lot of waste, 2) there is an opportunity for someone to do an easy piece of research."

Based on links from 2019-08-08 and 2019-08-09, we get one set of results (extracted all URLs saved from January 2019 through July 2019; http and https only, eliminated PDF and other links to obviously non-html material). Technical appendices and footnotes are below for those wishing to explore further and reproduce.







+ this only retrieves the first layer of javascript, far from all of it + some site gave wget trouble, should have fiddled the agent string, --user-agent="" + too many sites respond without proper HTTP response headers, slows collection down intolerably + the pages themselves often contain many dead links + serial fetching is slow and because the sites are unique

$ find . -mindepth 1 -maxdepth 1 -type d -print | wc -l 91 $ find . -mindepth 1 -type f -print | wc -l 4171 which is an average of 78 objects per "article"

+ some sites were tech blogs with lean, hand-crafted HTML, mainstream sites are much heavier, so the above average is skewed towards being too light

Quantity and size of objects associated with articles, does not count cookies nor secondary scripts:

$ find . -mindepth 1 -type f -printf '%s\t%p\n' \ | sort -k1,1n -k2,2 \ | awk '$1>10{ sum+=$1; c++; s[c]=$1; n[c]=$2 } END{ printf "%10s\t%10s\n","Bytes","Measurement"; printf "%10d\tSMALLEST\n",s[1]; for (i in s){ if(i==int(c/2)){ printf "%10d\tMEDIAN SIZE\n",s[i]; } }; printf "%10d\tLARGEST\n",s[c]; printf "%10d\tAVG SIZE\n",sum/c; printf "%10d\tCOUNT\n",c; }'

Bytes File Size 13 SMALLEST 10056 MEDIAN SIZE 32035328 LARGEST 53643 AVG SIZE 38164 COUNT









Overall article size [1] including only the first layer of scripts,

Bytes Article Size 8442 SMALLEST 995476 MEDIAN 61097209 LARGEST 2319854 AVG 921 COUNT

Estimated content [2] size including links, headers, navigation text, etc:

+ deleted files with errors or warnings, probably a mistake as that skews the results for lynx higher

Bytes Article Size 929 SMALLEST 18782 MEDIAN 244311 LARGEST 23997 AVG 889 COUNT

+ lynx returns all text within the document not just the main content, at 75% content the figures are more realistic for some sites:

Bytes Measurement 697 SMALLEST 14087 MEDIAN 183233 LARGEST 17998 AVG 889 COUNT

at 50% content the figures are more realistic for other sites:

465 SMALLEST 9391 MEDIAN 122156 LARGEST 11999 AVG 889 COUNT






       


$ du -bs * \ | sort -k1,1n -k2,2 \ | awk '$2!="l" && $1 { c++; s[c]=$1; n[c]=$2; sum+=$1 } END { for (i in s){ if(i==int(c/2)){ m=i }; printf "% 10d\t%s\n", s[i],n[i] }; printf "% 10s\tArticle Size\n","Bytes"; printf "% 10d\tSMALLEST %s\n",s[1],n[1]; printf "% 10d\tMEDIAN %s\n",s[m],n[m]; printf "% 10d\tLARGEST %s\n",s[c],n[c]; printf "% 10d\tAVG\n", sum/c; printf "% 10d\tCOUNT\n",c; }' OFS=$'\t'









[1]

$ time bash -c 'count=0; shuf l \ | while read u; do echo $u; wget --continue --page-requisites --timeout=30 "$u" & echo $((count++)); if ((count % 5 == 0)); then wait; fi; done;'









[2]

$ count=0; time for i in $(cat l); do echo;echo $i; lynx -dump "$i" > $count; echo $((count++)); done;








[3]

$ find . -mindepth 1 -maxdepth 1 -type d -print | wc -l 921

$ find . -mindepth 1 -type f -print | wc -l 38249









[4]

$ find . -mindepth 1 -type f -print \ | awk '{sub("\./","");sub("/.*","");print;}' | uniq -c | sort -k1,1n -k2,2 | awk '$1{c++;s[c]=$1;sum+=$1;} END{for(i in s){if(i == int(c/2)){m=s[i];}}; print "MEDIAN: ",m; print "AVG", sum/c; print "Quantity",c; }'









[5]

$ find . -mindepth 1 -type f -name '*.js' -exec du -sh {} \; | sort -k1,1rh | head 16M ./www.icij.org/app/themes/icij/dist/scripts/main_8707d181.js 3.4M ./europeanconservative.com/wp-content/themes/Generations/assets/scripts/fontawesome-all.min.js 1.8M ./www.9news.com.au/assets/main.f7ba1448.js 1.8M ./www.technologyreview.com/_next/static/chunks/commons.7eed6fd0fd49f117e780.js 1.8M ./www.thetimes.co.uk/d/js/app-7a9b7f4da3.js 1.5M ./www.crossfit.com/main.997a9d1e71cdc5056c64.js 1.4M ./www.icann.org/assets/application-4366ce9f0552171ee2c82c9421d286b7ae8141d4c034a005c1ac3d7409eb118b.js 1.3M ./www.digitalhealth.net/wp-content/plugins/event-espresso-core-reg/assets/dist/ee-vendor.e12aca2f149e71e409e8.dist.js 1.2M ./www.fresnobee.com/wps/build/webpack/videoStory.bundle-69dae9d5d577db8a7bb4.js 1.2M ./www.ft.lk/assets/libs/angular/angular/angular.js






[6] About page bloat, one can pick just about any page and find from one to close to two orders of magnitude difference between the lynx dump and the full web page. For example,




$ wget --continue --page-requisites --timeout=30 \ --directory-prefix=./test.a/ \ https://www.newsweek.com/saudi-uae-war-themselves-yemen-1453371 . . .

$ lynx --dump \ https://www.newsweek.com/saudi-uae-war-themselves-yemen-1453371 \ > test.b

$ du -bs ./test.? 250793 ./test.a 15385 ./test.b

Recent Techrights' Posts

The U.S. Patent and Trademark Office Hijacked Again by Patent Litigation Industry, as President Cheeto Prioritises Aggressors
The "mafia" has taken over the "industry" and the Federal system (justice and constitutions trampled upon)
Ubuntu Slop and FUD Manufactured With LLMs and Funded (by Oneself) 'Studies'
Slop and FUD are ruining the Web
Gemini Links 01/04/2025: Games and More
Links for the day
Why We're Reporting Brett Wilson LLP for Apparently Misusing Their Licence to Protect American Microsofters Who Attack Women
For those who have not been keeping abreast
 
Links 02/04/2025: Microsoft Developers Are Threatening to Go on Strike, World Backup Day Noted
Links for the day
Gemini Protocol Has Growing Appeal (the Web Got Too Bloated and Full of LLM Slop)
For any "data plan" with bandwidth limits or "tiers" it would be cheaper to use/browse Geminispace
The Web Can Survive LLM Slop, But Only If We Collectively Shun and Discourage Serial Sloppers
Doing nothing ought not be a possibility
Amid Secret Shut-downs and Mass Layoffs at Microsoft (4 Waves of Layoffs in 3 Months of 2025) Some Microsoft Staff Expected to Go On Strike
workers going on strike
Gemini Links 02/04/2025: No more on Mastodon and Gemini Mention Script in Go
Links for the day
Over at Tux Machines...
GNU/Linux news for the past day
IRC Proceedings: Tuesday, April 01, 2025
IRC logs for Tuesday, April 01, 2025
My Motion Disbarring or “Striking Off” Brett Wilson LLP for Enabling Violent Americans Who Try to Crush Microsoft Critics in the United Kingdom by Multiple SLAPPs
"Guns for hire" (for Microsoft people who received Microsoft salaries)
Links 01/04/2025: Apple Fined $162M for Privacy Abuses, Disinformation Online a Growing Concern
Links for the day
Newer Press Reports Confirm That Microsoft Shuts Down 'Hey Hi' (AI) Labs Despite All the Hype
The "hey hi" (AI) bubble is not sustainable
Links 01/04/2025: Mass Layoffs at Eidos and "Microsoft Pulls Back on Data Centers" (Demand Lacking); "Racist and Sexist" Slop From Microsoft
Links for the day
Stefano Maffulli and His Microsoft-Funded OSI Staff Are Killing the OSI and Killing "Open Source" (All for Money!)
This is far from over
Gemini Links 01/04/2025: XKCDpunk and worldclock.py
Links for the day
50 Years of Sabotage and a Gut Punch to Computer Science (and Science in General)
Will we get back to science-based computing rather than cult-like following?
Techrights Headlines as Semaphore
"If you are hearing this, thank you"
3 Months in 2025, 4 Waves of Mass Layoffs at Microsoft, Now Offices Shut Down Permanently
"A recent visit by the South China Morning Post confirmed that the office was dark, unoccupied, and had its logo removed."
Over at Tux Machines...
GNU/Linux news for the past day
IRC Proceedings: Monday, March 31, 2025
IRC logs for Monday, March 31, 2025
Links 31/03/2025: China Tensions, Bombs Falling in Myanmar After Earthquake
Links for the day
Gemini Links 31/03/2025: Falling Out of Love With Tech, Sunsetting openSNP
Links for the day
R.T.O. at IBM in Texas and Atlanta (State of Georgia) Expected as "Soft Layoffs" Catalyst This Coming Year
It also sounds like more IBM layoffs are in the making
Law Firms Can Also Lose Their Licence for Clearly Misusing It
The bottom line is, never made the false assumption that because you can pile up SLAPPs in a docket you will not suffer from bad reputation or even get disbarred
Link between institutional abuse, Swiss jurists, Debianism and FSFE
Reprinted with permission from Daniel Pocock
LLM Slop Piggybacking News About GNU/Linux and Distorting It
new examples
Links 31/03/2025: Press and Democracy Under Further Attacks in the US, Attitudes Towards Slop Sour
Links for the day
Open Source Initiative (OSI) Privacy Fiasco in Detail: The OSI Does Not Respect Anybody's Privacy
The surveillance mafia that bans dissent or key people (even co-founders) with dissenting views
Gemini Links 31/03/2025: More X-Filesposting and Dreaming in Emacs
Links for the day
Over at Tux Machines...
GNU/Linux news for the past day
IRC Proceedings: Sunday, March 30, 2025
IRC logs for Sunday, March 30, 2025