[tor-talk] Craigslist now giving Tor the slows, lol

grarpamp grarpamp at gmail.com
Thu Jun 12 05:43:50 UTC 2014


On Wed, Jun 11, 2014 at 11:35 PM, Mirimir <mirimir at riseup.net> wrote:
> Cool. What URLs have you been looking at -- root URLs or particular ads,
> posts, etc?

They're all the same timewise. Pick large pages, they
longer load times under delay blocking. The front page
is fine.

>> The division of half the relays is interesting.
>> If there's a better source for the probability each
>> exit will be picked by a client I'd like to know?

> You're testing all known exits, right?

Only ones with matching exit policy.

> And they're ranked in the plot by
> loading time, right?

Yes. There are 3-4 barely visible steps
interesting in the thick blocked line too.
I put 10min on the fetch, but those steps around
120-150 could also relate to some stack timeout
in the fetch - webserver path. Nor is the
test machine quiet.

> But in that case, I don't understand the zero
> weights in the center.
> I'd appreciate more explanation of this observation.

All as before, of the blocked exits, a sizeable fraction of
the ones around 120 sec had a consensus weight of
zero. No idea.

There also seems to be a topleft to bottomright
diagonal line weight density split.

> I'm guessing that Tor clients don't pick at least some of the
> non-responding ones.

Afaik they don't ...

> I'd also like a cite to the selection algorithm.

... torspec.git

> Also, how are you measuring load times?

However long it takes your fetch to return.

> Are you getting full pages with images etc, or just the HTML?

Reduce the variables... one page of html.

> What I'm wondering is whether Craigslist's distributed design and load
> balancing methods are slowing down access via Tor, but not intentionally.

Doubt it. But try tor-resolve and traceroute to see
what they're doing there. Afaik they are strictly in SFbay.

I can't tell more on the 50/50 time split without also, regarding
possible nagging abuse, at least checking the percent
of 'recently new ip' relays in the 1200, and probability of
being chosen exit from 1200 (I was trying correlatin weight
to blocking, but inconclusiv or i have wrong weight metric).


More information about the tor-talk mailing list