In a presentation at UCSD last week, the VP of technology at Facebook updated the public record on the number of servers Facebook is using. The last publicly available number was 10,000 but that number was assumed to be incredibly outdated given the rate of growth at Facebook. The new number is 30,000 servers but what I found to be even more incredible is that they are collecting 25TB of logging data/day.
I did some data center math and assuming that they are keeping a year of logging data (they are likely keeping far more), they need at least 70 racks of servers just to handle the storage of a years worth of logs. For my math, I assumed 4 servers of 24 hard disks per server per rack using 2TB disks and assumed some form of RAID; given the size of the individual disks, they have to use RAID 6 or 60 or something similar to protect against data losses as a result of disk failure. Even these numbers may be far too low. Western Digital’s RE4 enterprise level 2TB drive is quite new so if you change the assumption to the widely available 1TB drives and also change the assumption of 3 servers/rack, you end up with a number of more like 200 racks just for log data – assuming the logs are stored on hard drives. Since all of Oyster.com takes well under one rack, the Facebook numbers are mind bending.
