a few months ago twitter may have been interesting, mostly for the technical challenges. But now I'd rather hack on FriendFeed simply because it serves a supserset of the same purpose and is likely just as hard (in other ways) under the hood. But Paul, Bret, and friends have done an amazing job of hiding that so far.
funny, I was just thinking that. I'd love to work on twitter, not so much because I care about the service, but because it's fundamentally hard, and that's cool. Personally I wonder why someone hasn't already transitioned Twitter to a publish-subscribe architecture internally, and used a bunch of queues to buffer the impact of load spikes ( a slow tweet is not the same as a dropped tweet! ). But who knows if MySQL is really the best backbone for that anyways.
I will be interested to see where Zawodny does go. His tools have been worth a lot to me over the years, and he communicates well, so he ought to provide a ton of value to whatever his next gig may be.
Too bad for Yahoo. This and other high-profile departures may be coincidental, but rule #1 of a successful career is that you don't burn your bridges. So the kind words on exit don't really mean anything to me. I've pulled this sort of stunt before myself, as I swam away from a sinking ship whispering sweet nothings to the crew.
I'd love to work on twitter, not so much because I care about the service, but because it's fundamentally hard
Either I'm missing something, you're missing something, or we have different opinions of what "fundamentally hard" means. What's so hard about twitter? You can keep recent tweets in RAM and maintain on-disk per-user "all your friends' old tweets" files.
When I read "fundamentally hard", I think NP-hard. What is NP-hard about twitter? Am I missing something or is this a reference to a systems development issue?
Realtime and near-realtime is hard in practical terms, not mathematical terms. I work on the latter, and some days I miss working on the former.
If it were easy to get it right, and profitable to execute correctly, I have to imagine that a competitor would have done so by now. But given the time and resource constraints, it seems that the problem remains difficult to manage, and I've always enjoyed that sort of challenge.
Developing heuristics for increasing statistical power, robustness, or informativeness is a mostly cerebral pursuit. Operations is more bloody in-the-trenches mud wrestling. And sometimes I look out from the ivory tower and I miss it.
> I wonder why someone hasn't already transitioned Twitter to a publish-subscribe architecture internally, and used a bunch of queues to buffer the impact of load spikes ( a slow tweet is not the same as a dropped tweet! ).
As others have pointed out, it's not the messaging that's hard, it's the ad-hoc services-- friends-of-friends timelines, keyword search/subscriptions, etc. that are hard to scale and are what is bringing twitter down:
Yahoo is getting desperate. Have you noticed all the adds on the radio for jobs? I wonder who would want to jump in a sinking ship right now.
Unless they flush out most of the management fat, and political-cooporate overhead, and start from semi-scratch, they will never be an attractive place fore great people.
