Scalability is a data-analysis system’s ability to maintain or improve performance as data grows, queries become more complex, or more users and sources are added. It is not simply a matter of buying more hardware: the suitable approach depends on the workload, constraints, and performance goals. section 7.1 section 7.0
Assess scalability alongside the measures that reveal whether growth is harming the system:
- Throughput measures how many requests or tasks the system handles at once. section 7.1
- Bandwidth measures how much data the system imports or exports per unit of time. section 7.1
- Response time measures the complete duration of an analysis task, including computation, retrieval, and display. section 7.1
- Availability and reliability capture whether the system remains usable and produces accurate, timely results under changing conditions. section 7.1
Optimization before expansion
A Brazilian e-commerce analysis that once handled 100,000 orders may later process millions as the business becomes multinational. Improving the algorithm or pipeline is code optimization; adapting it to more CPUs, GPUs, or distributed execution is code scaling. A patient-monitoring unit may favor local optimization and high availability, while a financial-market system may preserve maintainable code and add capacity through parallelism, CPUs, or network bandwidth. section 7.0 section 7.2
Practical scaling options include parallel and distributed computing, GPU acceleration, specialized hardware or software, cloud deployment, and frameworks such as Dask. Scaling is not one-size-fits-all, and additional capacity does not replace measurement or retesting. section 7.0 section 7.5
