Zuck said that Facebook spends 3 percent of their CPU power on privacy. With such a low CPU budget dedicated to something as important as users' privacy, it's no wonder they do such a poor job of it.
Zuck said that Facebook spends 3 percent of their CPU power on privacy. With such a low CPU budget dedicated to something as important as users' privacy, it's no wonder they do such a poor job of it.
What would you consider to be a more reasonable amount of CPU budget to spend on excluding search results from some queries? I'm surprised it's as high as 10%, but I never really thought of CPU usage as a metric for privacy protection.
Maybe a simple notallowed() function doesn't scale linearly across many PB of data.
In a simple situation, you have a table or simple ruleset describing everything that is allowed, and disallow everything else. In that case, isallowed() is cheap. Or you might have a table of all the things that aren't allowed, which is dumb as a box of rocks but still cheap. FB must instead have rulesets where the operation of the rule depends on knowing the graph of relationships, and maybe also the provenance of the data which you're talking about (was it shared by route A or by route B?). Evaluating tha
%
APL is a natural extension of assembler language programming;
...and is best for educational purposes. -- A. Perlis
3 percent of CPU for privacy (Score:-1, Flamebait)
Re: (Score:5, Insightful)
Zuck said that Facebook spends 3 percent of their CPU power on privacy. With such a low CPU budget dedicated to something as important as users' privacy, it's no wonder they do such a poor job of it.
I read 10%, not 3%:
http://www.engadget.com/2013/01/15/facebook-graph-search/ [engadget.com]
What would you consider to be a more reasonable amount of CPU budget to spend on excluding search results from some queries? I'm surprised it's as high as 10%, but I never really thought of CPU usage as a metric for privacy protection.
Re:3 percent of CPU for privacy (Score:3)
How can 10% of a server farm go to that? if(notallowed(X,Y)) { etc
How is that notallowed() function written?
boggles my mind. Maybe I am alone and the Ubercoders at FB really can spend 10% of quality CPU time satisfying that func().
H.
Re:3 percent of CPU for privacy (Score:4, Insightful)
How can 10% of a server farm go to that? if(notallowed(X,Y)) { etc
How is that notallowed() function written?
boggles my mind. Maybe I am alone and the Ubercoders at FB really can spend 10% of quality CPU time satisfying that func().
H.
Facebook processes more than 500TB [slashgear.com] of data a day, and has over 100PB in its Hadoop cluster.
Maybe a simple notallowed() function doesn't scale linearly across many PB of data.
Re: (Score:2)
Maybe a simple notallowed() function doesn't scale linearly across many PB of data.
In a simple situation, you have a table or simple ruleset describing everything that is allowed, and disallow everything else. In that case, isallowed() is cheap. Or you might have a table of all the things that aren't allowed, which is dumb as a box of rocks but still cheap. FB must instead have rulesets where the operation of the rule depends on knowing the graph of relationships, and maybe also the provenance of the data which you're talking about (was it shared by route A or by route B?). Evaluating tha