The Tail At Scale
Blog Post by: Tong Chen, Kaitlyn Khan, Hamza Rashid, Tom Reingold In massive-scale systems that might be spread over a wide geographical area, response time for the user can be high. The statistical distribution of the latency is a long tail, i.e. as response time increases, the number of users experiencing a given response time decreases. This sounds OK, but the tail is long which is to say that a great many users can experience excessive response time. This paper discusses ways to mitigate long response time in ways that would not be necessary in smaller systems. (TR) Large companies such as Google divide large problems among many computers and networks, and these might even be continents apart. An example of a service would be auto-completion in the Google search bar. I might type “how do I” and the first three suggestions are - how can i tell what folder a google doc is in - how do i watch the super bowl - how do i get wordle (TR) The problem of guessing what the user wants is divi...