On: High Performance at Massive Scale: Lessons Learned at Facebook

[andrewcooke]

wow. they have no (ie random) partitioning - it's just brute force. i hope their "Graph-based caching and storage systems" is coming along... :o)

[edit: although, given the exponential growth they give and the factor of 4 headroom they have left, they're going to top out in about 1.5 years anyway]

[niels → andrewcooke]

Yep, so basically one request can touch hundreds of servers. It's basically the opposite approach of Twitter. They aggregate the graphs on write.

[aaronblohowiak → niels]

Who is "They" in this case? Both a read and a write could be a "request", so I am having some trouble with the ambiguity of your last sentence. Thanks!

[niels → aaronblohowiak]

Twitter is they.

[niels → niels]

Which also is the reason you can only have 5000 friends on FB. If you had a million friends, you'd probably touch all their servers for one request.