Skip to content
Omnikalc

Universal scalability law calculator

Enter the contention and coherency penalties, the number of nodes and the single-node throughput to get the speedup, the throughput, the efficiency and the number of nodes at which adding more nodes stops helping.

Updated on
Your inputs
%

Value out of range

%

Value out of range

Value out of range

requests/s

Value out of range

Results

Speedup with n nodes

8.04

Throughput
804 requests/s
Efficiency
50.3 %
Nodes at the throughput peak
30.8
Speedup at the peak
9.04
Ceiling from contention alone (1/α)
20.0
Nodes left before the peak
14.8
Adding nodes still increases the throughput

The universal scalability law calculator gives the speedup, the throughput, the efficiency and the number of nodes at the peak of a system with a contention penalty α and a coherency penalty β.

How the calculation works

  • Enter the contention (α) and coherency (β) penalties as percentages.
  • Enter the number of nodes and the throughput of one node.
  • The calculator gives the speedup, the throughput, the efficiency, the peak node count, the peak speedup and the ceiling from contention.

Example

With α = 5 %, β = 0.1 % and 16 nodes, the denominator is 1 + 0.05 × 15 + 0.001 × 16 × 15 = 1.99, so the speedup is 8.04 and the efficiency 50.3 %. At 100 requests per second per node, that is 804 requests/s. The throughput peaks at √(0.95 / 0.001) = 30.8 nodes, with a speedup of 9.04, and Amdahl law alone would cap the speedup at 20.0.

Why it matters

The calculator helps performance engineers, architects and students who plan the scaling of databases, application servers or parallel programs and want to know where adding hardware stops paying off.

Good to know

The law is empirical: α and β summarize many effects, such as locks, queues, network traffic and cache invalidation, and are best obtained by fitting measurements. A β of zero gives the Amdahl curve, which only flattens. Real systems may have a linear region before the knee. The efficiency tells how much of each node’s capacity is useful at the chosen size. If a system is already past its peak, reducing β, for example with less shared state, brings more than adding hardware. The throughput prediction assumes the workload per node stays the same, and the measurement of the single-node value matters a lot.

The Amdahl law calculator covers the contention-only case, the server capacity calculator sizes servers, the queue wait time calculator covers waiting lines and the reserved instance calculator compares cloud costs.

Frequently asked questions

What does the law say?

The relative capacity of a system with N nodes is C(N) = N / (1 + α (N − 1) + β N (N − 1)). The α term is contention for a shared resource, as in Amdahl law, and the β term is the cost of keeping the nodes coherent, which grows with the square of N.

Why can throughput fall with more nodes?

With a coherency penalty, each added node must exchange data with all the others, and this cost eventually outweighs the benefit of its capacity. Beyond N* = √((1 − α) / β), the throughput decreases, which Amdahl law cannot show.

How are α and β found?

Measure the throughput at several load levels, say 1, 2, 4, 8 and 16 nodes, then fit the curve to the formula by regression. The coefficients are properties of the system and workload, so they should be refitted after major changes.

Was this calculator helpful?

Thanks for your feedback!

Results are estimates provided for information purposes. They do not replace professional advice.