Home
entries friends calendar user info My Google Pages Previous Previous Next Next
Brett Morgan - The science behind Twitter
domesticmouse
[info]domesticmouse
Add to Memories
Tell a Friend
The science behind Twitter
I've started on a journey of understanding the architecture behind Twitter. I mean, I understood some of the adventure Twitter had gone on to get past their Fail Whale period, but now I am reading through the papers that Evan Weaver has been using as part of their rebuild. So many names I recognise from previous forays into the Computer Science paper stack. Werner Vogels, Leslie Lamport, Barbara Liskov, Jim Gray and Michael Stonebreaker. A bit of a who's who of the establishment, really.

I'm intrigued to see what happens next.
Comments
thorfinn From: [info]thorfinn Date: July 6th, 2009 07:46 am (UTC) (Link)
The funny thing about twitter is that I suspect the fixes are pretty much a re-imaging of the IRC protocol in that context. Those who fail to learn about history are doomed to repeat it.

Good to know that they've got around to learning about history (i.e., actually looking at Computer Science research in the area).
domesticmouse From: [info]domesticmouse Date: July 7th, 2009 12:18 am (UTC) (Link)
IRC, if i recall correctly, is a message distribution system with constantly connected clients.

Twitter is heading in the direction of being a message distribution system - that is the core of the SEDA design pattern - but it has the additional wrinkle that the clients connect sporadically. Thus, it is actually a harder problem to solve than the IRC problem.
thorfinn From: [info]thorfinn Date: July 7th, 2009 02:23 am (UTC) (Link)
No, it really isn't a harder problem. IRC clients connect and disconnect all the damn time - that's the whole point. IRC *servers* connect and disconnect from each other all the time too. Nobody spends their whole life connected to IRC. It's a full on distributed message system with randomly connecting and disconnecting clients. Really.

The wrinkle with twitter is the logging and archive-retrieval aspects (which IRC doesn't have), but that is nowhere near a problem if you accept that you can do it in asynchronous time for readers, rather than synchronous time.
domesticmouse From: [info]domesticmouse Date: July 7th, 2009 05:49 am (UTC) (Link)
Mmm, to properly discuss this I need a whiteboard.
4 comments or Leave a comment
profile
Brett Morgan
Name: Brett Morgan
calendar
Back November 2009
1234567
891011121314
15161718192021
22232425262728
2930
page summary
tags