2009/3/31 Michael DeHaan <mdehaan(a)redhat.com>:
Max Spevack wrote:
> On Tue, 31 Mar 2009, Wojtek Walczak wrote:
>
>
>> OK, that's all for now, thanks for reading :-)
>>
>> All your comments are greately appreciated!
>>
>
> This is all very, very interesting, and I admire and appreciate the time
> that you took to put these ideas together.
>
> On a high level, my thought is this:
>
> We need to be smart about how we extend EKG's functionality. What do I
> mean by this? I mean that the "sources" and "facts" -- the
actual data
> from projects, that is immutable -- need to be managed cleanly and
> simply. That was one of the requirements that I put on Yaakov when he
> started the v3 rewrite.
>
To expand on what Max has said...
My ongoing theme is "let's not collecting data and build graphs for data
we can't act on".
For instance, what users talk to what users is probably not something we
can act on ... but knowing things about the overall
pulse of different communities is.
The idea is, how can we use that data to improve Fedora or learn what to
invest in or what people care about.
Graphs and such are seductive, so we need to make sure we're spending
time on the right ones. (Hence, why I think IRC
is not an interesting demographic -- the data associated with it is not
very rich and the cost of mining it is higher)
Doing simple things first, like being able to compare one project
against each other is more important than finding out what
the network of associations between individual contributors is ... or
even expanding it to new data sources before it's ready.
For instance, the level of metadata provided by, say, forums, is rather
inferior to that provided by mailing lists (no domains), so
we are likely to end up with islands of data that may be less actionable
than the other data.
On the other hand, commit data is pretty rich -- we know the size of the
deltas and also users+domains.
Email addresses -- not being provided by forums -- makes them less
interesting.
That's kind of general rambling but I guess what I'm saying is (A) let's
go after data that is already strongly connected to other data or
keys that we have, and (B) let's seek to find data that helps answers
new questions and solve problems, rather than just being conceptually
interesting (trying to figure out who replies to who doesn't achieve this).
Agreed.
-Yaakov