Skip to content

Experimental features

gorhill edited this page Oct 3, 2014 · 44 revisions

This page is about the experimental features which you can enable in µBlock. Features which I think are good for the users but which I consider aren't yet mature in design/implementation are considered experimental.

Experimental features are disabled by default, you can enabled them in the Settings tab in the dashboard. As long as a feature is labeled "experimental", I can't guarantee it will work flawlessly.

Privacy exposure reduction: local mirroring

One way users increase their privacy exposure is through content delivery networks ("CDN") and likes. Good examples of CDN are ajax.googleapis.com, cdnjs.cloudflare.com, googletagservices.com, etc. These servers are used to serve resources to countless web sites.

When a resources is pulled from one of these CDNs, typically the referrer header is set in the HTTP request, which allows ubiquitous CDNs to collect data about your browsing habits. Not good for privacy.

In the spirit of reducing privacy exposure, local mirroring is introduced as an experimental feature in µBlock. Local mirroring allows µBlock to cache locally resources pulled from known ubiquitous CDN, and future requests for the same resource will be served locally rather than pulling it from the CDN: not pulling a resource from a CDN prevents that CDN from collecting data about your browsing habit.

zdnet.com

The picture above shows that connections to ajax.googleapis.com, www.googletagservices.com, twitter.com were prevented -- hence no trace left in their server logs), and local copies of the requested resources were served instead -- hence no page breakage. (Surprisingly, the above shows that requests to googletagservices.com are not blocked when using EasyList + EasyPrivacy. ("Dan Pollock's", MVPS and hpHosts block it though.)

A quick benchmark -- using reference benchmark, with the feature disabled vs. enabled shows the following:

Disabled:
URLs visited: 60
Domains: 415 / 475
Scripts: 857 / 1264
Outbound cookies: 0 / 130
Net requests: 3,304 / 6,264

Enabled:
URLs visited: 60
Domains: 337 / 397
Scripts: 793 / 1214
Outbound cookies: 0 / 132
Net requests: 3,174 / 6,156

As seen above, a significant amount of connections to third-party ubiquitous CDNs were foiled with local mirroring enabled. This contribute to reducing privacy exposure, without breaking web pages.

Local mirroring results

Advantages of local mirroring:

  • Reduction of privacy exposure
  • Less bandwidth
  • Faster page load

Disadvantage of local mirroring:

  • Higher memory consumption
Update

[Following cut & pasted from here]

I ran a new benchmark with uBlock 0.6.6.0-rc.0 and ABP 1.8.5 with the same lists (except Peter Lowe's because there is no easy convenient way to install it in Adblock Plus -- so I assume users are likely to not bother).

Here is the resulting diff: https://www.diffchecker.com/5z91i47m

In red what ABP blocked which was not blocked by uBlock. In green what uBlock blocked which was not blocked by ABP.

I suspect the only two requests not blocked by uBlock which were blocked by ABP is just the result of the page content changing between the time I benchmarked uBlock and ABP.

Filter lists:

  • ABP: Out of box settings + EasyPrivacy, Malware domains, Fanboy's Social Blocking List, Anti-Facebook + "Acceptable ads" disabled
  • uBlock: Out of box settings + Anti-Facebook + Local mirroring enabled and primed (as it would be through normal use).

What local mirroring accomplished in the above benchmark... Prevented connections to:

  • 2mdn.net
  • ajax.googleapis.com
  • cloudflare.com
  • fonts.googleapis.com
  • googletagservices.com
  • gstatic.com
  • janrain.com
  • jquery.com
  • twitter.com

And since the remote resources were available locally, preventing the connection didn't cause page breakage.

Clone this wiki locally