[tor-dev] Scaling tor for a global population

[Mike Perry]

Sebastian Hahn:
> 
> On 27 Sep 2014, at 02:18, Mike Perry <mikeperry at torproject.org> wrote:
> > If we were willing to tolerate 10% directory overhead this would allow
> > for 5 times as many users. In other words, 100M daily connecting users.
> > 
> > We would still need to find some way to fund the growth of the network
> > to support this 40X increase, but there are no actual *technical*
> > reasons why it cannot be done.
> 
> One thing there isn't an automatic answer to is whether our current users
> and our would-be users are differ in their usage pattern. Currently, my
> intuition would be that most of our users are responsible for a relative
> small amount of traffic, whereas we have some users who pull a lot of
> data. I wonder what happens with Netflix, youtube and other services.
> We might at least want to try and figure this into the equation by
> estimating our average daily bytes sent/received per users, and
> comparing that to the bytes sent/received by our target group.

This is a great point, Sebastian. Some of the browser vendors do this
sort of analytics, and other parties have also made some inferences
independently.

In particular, these two studies are old, and may not be accurate
anymore, but they say that between 5-15% of each browser's userbase
used Private Browsing Mode, depending on the browser (pg 9):
https://www.usenix.org/legacy/event/sec10/tech/full_papers/Aggarwal.pdf

And moreover, that the typical usage duration for Private Browsing Mode
was 4-22 minutes, with 10 minutes being most common:
https://blog.mozilla.org/metrics/2010/08/23/understanding-private-browsing/

Refreshed studies of this type of data would be very helpful for
determining what size rollout we could handle, and if we'd want to
further limit it by making the Tor mode opt-in, not prominent, or
otherwise less likely to be selected by most users, at least for the
initial versions.

We could also capture more detailed usage frequency analytics and
bytecount statistics in an alpha or other trial version.


-- 
Mike Perry
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 801 bytes
Desc: Digital signature
URL: <http://lists.torproject.org/pipermail/tor-dev/attachments/20141002/6bf6a27f/attachment.sig>