> The biggest one, in my mind, is the 'one POST per follower per message' problem.
Yeah, all distributed systems must communicate with their peers. In the worst case this means sending a POST for each message to each subscriber since each subscriber is a different server. This can be optimized by pipelining messages that were sent within the same time window.
In the best case, which is probably the most common, multiple users will share the same host and the protocol can be aware of this and add an envelope that specifies all subscribers on the host with a single copy of the message sent to each host instead of each subscriber. We plan to add this optimization before Tent 1.0.
Re:App.net State of the Union
by steveklabnik on Wednesday May 07, 2014 @01:04PM
(#7711342)
Right. This is the kind of thing you'd basically get for free if you just used PuSH instead of reinventing the world.
Anyway, good luck. My efforts in this space have failed and you're still plugging away, so...
Awesome, great.
The biggest one, in my mind, is the 'one POST per follower per message' problem. The previous stance was, and I realize I'm being a little bit uncharitable with this characterization, "we want people who use the service a lot to have to pay, so we're keeping the protocol inefficient for this purpose." Is this still the way things work?
And yes, while a lot of them do come down to opinion, a service that re-invents the world in this space sends off really bad signals. It's just a different kind of lock-in. Same beef I had with App.net.