Web service scalability

Scalability is constrained by contention and crosstalk. The time when your system is unable to keep up with load might come unexpectedly 🚒. Earlier you know your system's limits, better you're prepared to handle increasing load (optimize, redesign). This presentation provides an overview of how to measure and estimate web service's scalability. It is inspired by Baron Schwartz's talk Scalability is Quantifiable.

Run present to view the slides with notes:

$ git clone https://github.com/marselester/scalability.git
$ cd ./scalability/
$ go run golang.org/x/tools/cmd/present -notes

Universal scalability law

Forecast how concurrency N (nodes count) affects throughput X (requests per second). The example is taken from Practical Scalability Analysis with the Universal Scalability Law book.

You can paste usl.r content in R interactive session or execute commands with source:

$ brew install r
$ r
> source("usl.r")
> lambda
  lambda
995.6486
> sigma
     sigma
0.02671591
> kappa
       kappa
0.0007690945

The system's throughput will increase up to 35 nodes:

N = sqrt((1-σ) / κ) = sqrt((1 - 0.02671591) / 0.0007690945) = 35

Peak throughput will be 12,341 requests per second:

X(N) = (λ * N) / (1 + σ * (N-1) + κ * N * (N-1)) =
(995.6486 * 35) / (1 + 0.02671591 * 34 + 0.0007690945 * 35 * 34) = 12341