Glossary
Cloud scalability
Cloud scalability is a system's ability to absorb growth in data, users and workload by adding resources while performance and cost per unit stay acceptable. For storage it is judged less by how much capacity can be added than by what stays steady as capacity, object count and traffic increase.
Scalability, elasticity and the two ways to grow
| Property | What it describes | Where it matters most |
|---|---|---|
| Scale-up | Adding drives behind the same controllers | Small systems; ends when controllers saturate |
| Scale-out | Adding whole servers, each with disks, processors and network | Large data stores that keep growing |
| Elasticity | Growing and shrinking quickly to follow a short peak | Compute that follows seasonal or daily demand |
Scale-up reaches a point where capacity keeps rising while performance stops, so cost per unit of throughput climbs. Scale-out storage adds capacity and performance together. Elasticity is a separate property: storage rarely shrinks, so for data it is sustained growth that counts, and a platform can be highly elastic for compute and still scale poorly for data.
Limits a pilot never reaches
Systems that run well at a few hundred terabytes can stall at ten petabytes without running out of raw capacity. Three limits appear only at size:
- Object count. Billions of small objects strain metadata, listings and indexes well before the disks are full.
- Rebuild time. A 20 TB drive written at a sustained 100 MB/s takes over two days to rebuild with no competing traffic. At petabyte scale drives fail routinely, so rebuilds overlap with each other and with production load.
- Rebalancing. New nodes trigger data movement to even out load, and that traffic competes with applications for the same network and disks.
A test at 5 percent of target size says little about behavior at full size. Protection schemes such as erasure coding, which spread rebuild work across many servers, change how recovery behaves as the system grows.
Measuring it
Scalability shows as three lines plotted against capacity and object count: throughput, latency and cost per terabyte. Flat lines mean the system scales; a bend shows where the next limit sits. The workloads that need the measurement most are those whose data only grows: unstructured files, backups and AI datasets, covered further under petabyte storage.
Frequently asked questions
What is the difference between vertical and horizontal scaling?
Vertical scaling makes one machine larger; horizontal scaling adds more machines. In storage these correspond to scale-up and scale-out.