Re: user surveys for the Seul project

[Roger Dingledine <arma@mit.edu>]

In message <35895A1C.321FE71B@ix.netcom.com>, kmself@ix.netcom.com writes:
>Ok.  This is a way of looking at things I hadn't considered.  My
>original idea was screenshots and videos, plus system info.  Your focus
>is system info, and probably no video.  I can see just running a
>datarecorde and studying the results as being something much easier to
>do on a large scale.  I *don't* see statements about privacy or control
>of the product in your post.  I really do want to hammer this point --
>any watching/reporting device MUST be used in a very controlled, very
>open, and very user-aware manner.

I guess I was picturing that a bunch of people from freshmeat or slashdot
would grab it and install it because they wanted to help out the linux
movement. It would show them what it was going to transfer, and they could
accept or reject the submission. I think privacy is less of an issue when
it's all statistical anyway -- I agree that we need to be careful about
what data we collect, because privacy is a big issue. But as long as we're
not actually doing any screen captures, and all reported data is in the
form of "he had 5 windows open then, but he wasn't typing anything and
his cpu was idle", then everything is fine. Even if it's stuff like "when
he's using Netscape he often does task-switches to his xterm and back" or
"he often spends a lot of time idle when his mouse focus is on his
wordprocessor", I think that's fine. But yes, we should be aware of privacy
issues.

>OTOH, we seem to be getting toward the realm of an advanced system
>admin/analysis tool -- some of this info would come directly from system
>logs, much of the rest from /proc, some from utilities similar to
>existing X monitoring tools.

Yep. If we do it right, it would be a very extensible project. But it
would take a good chunk of work to do right.

>> There are no videos to look at, and no question of how to get the hardware
>> in place and set up the experiment and stuff. You just get people who want
>> to help out to run this daemon in the background, and voila, it keeps track
>> of things.
>
>Don't totally discount video.  Timestamped, periodic snapshots could put
>a lot of context on system activity.  Small scale => good.

Small scale means specific data and lots of data per person. Large scale
means not much specialized data per person, but in the long run it gives
us the big picture. I think either one can work. I'm looking at large
scale as a means of saving energy on the interpretation side of things.

(Ack. But I said we should work on the survey first. Really. :)

--Roger