…at least, that’s what they claim in a
new article about their architecture.
Huh.
Don’t get me wrong, it’s definitely big, but I would have guessed at least a few times more, if not a whole order of magnitude.
Anyway, #TIL and all that. Learn something new every day!
The part that stood out to me was “6000 requests per second are spent on writes”.
BTW, that article was sourced from a talk by Raffi Krikorian. You can watch it with slides at http://www.infoq.com/presentations/Twitter-Timeline-Scalability
well, that’s just for reads on timelines. that doesn’t include tweet writes, retrieval, search, etc.
crap, i knew an expert would call me on my BS eventually. :Pagreed, it’s a simplification, but it sounds like a fair one. based on my reading,+Raffi Krikorian is saying that the 300K number includes both search and timeline reads: “At 150 million users with 300K QPS for timelines (home and search) naive materialization can be slow.”writes are definitely much heavier, but looking naively at just volume numbers, they’re only 6kqps, ie just 2% more.regardless, point taken, my tone was probably off. i didn’t mean it as “ha ha, twitter’s so small” so much as “wow, if they’re one of the most meaningful internet services in the world, with as many users as they have, this means they’ve done some decent optimization on traffic volume and shaping. kudos to them.”
Whoops- missed the part about search timelines being included (I thought it was only home and user).It’s amazing what caches can do. :-)
I also expected the number to be bigger.
via plus.google.com
+1ed this.
via plus.google.com
+1ed this.
via plus.google.com
+1ed this.
via plus.google.com
+1ed this.
via plus.google.com
reshared this.
via plus.google.com
reshared this.
via plus.google.com
Marco Valerio Barbera mentioned this Article on plus.google.com.