Two of my favorites are the Universal Scalability Law and Little’s Law, which are both in the territory of queueing theory and apply to modeling the throughput of software systems… and people systems. There’s a good explanation of both (applied to social contexts) in Adrian Coyler’s notes on the paper “Applying the Universal Scalability Law to Organizations”.
Little’s Law describes the average length of a queue, given the average rate at which new work items arrive and the average time to handle a single item.
people waiting in line = arrival rate * time to complete one transaction
The Universal Scalability Law describes how the throughput of work items scales with the number of workers handling items. This one is a little more complicated, but I’ll break it down.
C(N) = N / ( 1 + alpha * (N-1) + beta * N * (N-1) )
C is the capacity (maximum possible throughput of the system), N is the number of workers, alpha is a measure of contention (how much workers are blocked trying to get access on shared resources), and beta is a measure of coherence (how often workers have to communicate in order to reach agreement on what to do). Both alpha and beta range from 0 to 1.
When both alpha and beta are 0, each worker can just crank through the items in their own line without any slow-down. Perfect scalability!
When you have contention (alpha > 0), workers have to wait on access to a central resource before they can complete a work item. Imagine a series of grocery store checkouts, with one manager supervising. When two clerks need help from the manager at the same time, that’s contention. If every transaction requires the manager, then that’s alpha = 1 and there’s no benefit from having multiple clerks.
Coherence costs (beta > 0) happen when the workers have to come to consensus about something in order to proceed. To understand coherence, imagine that none of the clerks in the grocery store can recognize which vegetable is which. Whenever a shopper is buying a vegetable, the clerk handling that transaction calls all the other clerks over to vote on whether or not the vegetable in question is broccoli.
Coherence costs can hurt worse than contention because they can make multi-worker performance worse than the single worker case.