What's slowing down my team?

Overview

In my experience, team performance is often discussed in fairly ad-hoc terms. We have plenty of formalisms to reason about network communication and scalability, why not apply some of these to team performance?

All we need is some way to measure work performed and a unit of concurrency.

Agile teams have both of these. Team size is our unit of concurrency and Story Points is the way we can measure work.

Modeling team performance

1. Story points per sprint
2. Number of people in the team

The total number of points that will be burned down in a Sprint is,

• The average number of points for one person
• Times the number of people in the team

Work per Sprint - Expression

$$ P(people) = storypoints * people $$

$$ P(people) = \frac{storypoints * people}{1} $$

Impediments to linear scalability

• Contention - Usage of a shared, finite resource
  • Usually a delay in starting or continuing work
• Crosstalk - A form of direct or indirect communication
  • Usually an interruption of work

Contention

Example : One person runs the tests

Contention

Contention - relation to scalability

$$ P(people) = \frac{storypoints * people}{1 + (contention * (people - 1))} $$

Contention - Example

Crosstalk

A form of direct or indirect communication between people. Also known as coordination or cohesion. Generally an interruption of work. For example : Daily standup

Eveybody communicates with everybody else so the number of channels for a groups is:

$$ Channels(people) = people * (people - 1) $$

$$ Channels(people, talkers) = talkers * (people - 1) $$

Crosstalk - channels

Crosstalk - channels

Crosstalk - full mesh

Crosstalk - crosstalk channels

Crosstalk - relation to scalability

$$ f(x) = \frac{nx}{1 + kx(x-1))} $$

• n is concurrency - ideal parallelism, linear scaling
• k is cohesion - time waiting for data to become coherent

Crosstalk - example

Black box models - Team

One person runs the test suite	One team runs the load tests
Isomorphism

Black box models - Organization

Universal Scalability Law

$$ USL(x) = \frac{nx}{1 + \sigma(x-1) + \kappa x(x-1)} $$

• n is concurrency - ideal parallelism, linear scaling
• s is contention - cost of sharing resources
• k is cohesion - time waiting for data to become coherent

Interactive exploration

https://www.desmos.com/calculator/ekkdy5mlhc

DevOps vs Command and Control

Variability

Tasks vary in size and complexity.

Leads to variability in,

• Compute (CPU)
• IO (Network)
• Storage (RAM)

Further reading

• How to quantify scalability
• usl4j and you
• usl4j github
• LISA17 - Scalability is quantifiable
• Nelson Rules