Skip to content
nonconvexcard

Fit a scalability curve and find where it turns over

NumericalAdvanced⏱ 1 hourAI Systemsuniversal scalability lawcontentioncoherencycapacity

Problem

A distributed index reports relative capacity C(p)C(p) at pp shards:

pp 1 2 4 8 16 32
CC 1.00 1.96 3.72 6.63 10.08 11.39
  1. Fit C(p)=p/(1+α(p−1)+βp(p−1))C(p) = p/(1+\alpha(p-1)+\beta p(p-1)) by least squares on p/C(p)−1p/C(p) - 1, which is linear in the two parameters with no intercept.
  2. Predict p⋆p^\star and the capacity there.
  3. The team plans to run 64 shards. What does the model predict, and what would you tell them?

Compute the requested quantities and say what they mean. A number without an interpretation is not an answer.

Hints

Hints — 0 of 2 revealed

Solution

Opening this before you have a real attempt on paper is the fastest way to feel like you understand something you cannot reproduce.

Related problems

All problems
SolvedProblemFieldTypeDifficultyTime
Apply Little's law to a serving targetLLM OpsProofIntermediate30 min
How many users fit on one GPULLM OpsNumericalIntermediate30 min
What fitting random labels rules outDL TheoryProofIntermediate30 min