Net filtering: not worth to use bits of domain name in category hash [performance]

[gorhill]

No description provided.

[gorhill]

Results:

µBlock> onBeforeRequest: 0.108 ms (20488 samples)
µBlock> onBeforeRequest: 0.108 ms (20702 samples)
µBlock> onBeforeRequest: 0.108 ms (21068 samples)
µBlock> onBeforeRequest: 0.107 ms (21335 samples)
µBlock> onBeforeRequest: 0.107 ms (21487 samples)
µBlock> onBeforeRequest: 0.107 ms (21634 samples)
µBlock> onBeforeRequest: 0.108 ms (21808 samples)
µBlock> onBeforeRequest: 0.108 ms (21867 samples)
µBlock> onBeforeRequest: 0.108 ms (21984 samples)
µBlock> onBeforeRequest: 0.108 ms (22114 samples)

[gorhill]

Actually, the SpecificParty buckets are not even needed: the logic for that one is in the filters, as opposed to 1st-, 3rd-party, and types. Removing the SpecificParty buckets:

µBlock> onBeforeRequest: 0.089 ms (19684 samples)
µBlock> onBeforeRequest: 0.089 ms (20017 samples)
µBlock> onBeforeRequest: 0.088 ms (20405 samples)
µBlock> onBeforeRequest: 0.088 ms (20674 samples)
µBlock> onBeforeRequest: 0.088 ms (21033 samples)
µBlock> onBeforeRequest: 0.088 ms (21336 samples)
µBlock> onBeforeRequest: 0.088 ms (21491 samples)
µBlock> onBeforeRequest: 0.088 ms (21685 samples)
µBlock> onBeforeRequest: 0.088 ms (21816 samples)
µBlock> onBeforeRequest: 0.088 ms (21965 samples) 

[ghost]

Wow! Those figures are only half of the time compared with the ones you published here. That's fantastic - µBlock's lead over ABP is getting bigger and bigger! Congratulations, Raymond!

[gorhill]

Thanks.

The case here is a good illustration of what I said a while ago in the last paragraph in there (many other things in the article are outdated though, or even plain wrong as per findings of more benchmarking a while ago).