1 - About

Performance has two dimensions (two metrics):

Performance is a feature.

It can sometimes difficult to talk about performance when one technique improves throughput but decreases response time.

Which has the best performance ?

Vehicle Time to Paris Speed Passengers Throughput (pk/h)
Ferrari 2 hours 250 km/h 2 500
Greyhound 5 hours 100 km/h 60 6000

This two metrics are a subset of Service Level Indicator metrics.

3 - "Time to do the task" versus "Task par unit of time"

Vehicle Vehicle Ferrari Greyhound
Speed 160 mph 65 mph
Time to Bay Area 3.1 hours 7.7 hours
Passengers 2 60
Throughput (pm/h) 320 3900
  • Time is measured in time units/job.
  • Throughput is measured in jobs/time unit.

Then <MATH>\text{time} = \frac{1}{\text{throughput}}</MATH>

The formula is only valid for one job at a time, It takes 4 months to grow a tomato. Can you only grow 3 tomatoes a year ??

4 - Rule


  • data movements (increase spatial localization). storage: data next to the CPU (see Memory - Hierarchy). RAM better than Disk better than NAS ….
  • data operation (See Performance - Algorithm (Asymptotic analysis))
  • data volume (selectivity, less IO)
The bucket list of performance gains:
  • don’t do it
  • do it only once
  • do it less often
  • do it more efficiently

Performance is a funny thing. You can't make computers faster, you can only make them:
  • do less wasted work
  • or spend less time waiting.

5 - Example

A data system may drop some data of a query in order to be able to calculate results in a reasonable time using a reasonable amount of computing resources

6 - Quote

Within performance issues, encoding/decoding is often the most significant cost.

7 - Documentation / Reference

Data Science
Data Analysis
Data Science
Linear Algebra Mathematics

Powered by ComboStrap