Glossary
Storage QoS
Storage QoS, short for quality of service, is a set of controls that divide a shared storage system's performance among workloads by capping, reserving or weighting it. It exists so that one busy application or tenant cannot push everyone else's response times out of range.
What problem does QoS solve?
On shared storage, one tenant starting a batch job can multiply everyone else's response times. This is the noisy neighbor problem. A device moving from 40 to 85 percent busy answers requests several times slower, as covered under storage latency, so the damage spreads well beyond the workload that caused it. Environments that host many teams, as in multi-tenant storage, hit this first.
Which control does what?
| Control | What it does | What it costs |
|---|---|---|
| Limit (cap) | Stops a workload exceeding a set rate | Wastes performance when the system is otherwise idle |
| Reservation (floor) | Guarantees a minimum rate | Locks up capacity even when unused |
| Weight | Splits performance in proportion when workloads compete | Gives no fixed guarantee |
| Priority class | Lets important requests go first | Can starve lower classes |
| Latency target | Throttles others when a protected workload slows | Needs good measurement to work |
Each trades fairness against efficiency differently. Hard caps are simple and protect neighbors, but they hold back a workload even when nobody else needs the performance. Floors suit latency-sensitive applications. Weights suit the rest, since they share spare capacity and only enforce proportions during contention.
What tends to be forgotten?
Internal work competes for the same drives and links as applications. A rebuild after a drive failure, replication to another site, and data movement between tiers all consume bandwidth. Left outside QoS, they take performance from applications at the worst moment, and applications taken at full speed slow rebuilds, which prolongs the exposure described under storage high availability. Placing background work under the same policy keeps both within bounds.
QoS divides what exists. It does not create it, so a system that is short of IOPS or throughput overall is still short after limits are set. Sizing and storage performance measurement come first.
Frequently asked questions
What is the noisy neighbor problem?
One workload on shared storage takes so much performance that others slow down. QoS limits or weights are the usual way to contain it.
Are limits or reservations the better control?
Limits protect neighbors but waste idle performance. Reservations protect a sensitive workload but lock up capacity. Many environments use floors for a few critical applications and weights for the rest.
Does QoS apply to object storage?
Yes. Shared object storage serves many applications and tenants, and request rate or bandwidth controls per tenant keep one from consuming the shared capacity.
Related terms
- Multi-tenant storage: the shared setting where QoS matters most.
- Storage latency: the cost of contention.
- IOPS: a common unit for limits.
- Storage throughput: bandwidth that QoS divides.
- Storage high availability: why rebuild traffic needs control.
- Storage pooling: the shared resource QoS divides.