# Performance & Tiering

A dedicated cluster lets you match device classes and data-protection schemes to each workload, rather than accept a one-size-fits-all tier.

## NVMe as a fast tier or cache

You use NVMe drives two ways:

- As a dedicated fast pool. Place latency-sensitive workloads (databases, metadata, hot buckets) on an all-NVMe pool, and bulk or cold data on high-capacity drives.
- As a cache layer for spinning disks. Front high-capacity HDDs with NVMe so frequently accessed data is served from flash.

A modest amount of NVMe paired with high-capacity drives delivers strong performance per GB stored, without paying for all-flash capacity.

## Replication vs erasure coding

Per pool, pick the data-protection scheme for the workload:

- Replication (for example 3x). Lowest latency and simplest recovery. Higher raw-capacity overhead.
- Erasure coding. Much better usable-to-raw capacity ratio. Best for large, throughput-oriented or archival data.

Mixing schemes across pools in the same cluster lets you tune cost and performance per workload.

| Resource | URL |
| --- | --- |
| Ceph pools | [https://docs.ceph.com/en/latest/rados/operations/pools/](https://docs.ceph.com/en/latest/rados/operations/pools/) |
| CRUSH / device classes | [https://docs.ceph.com/en/latest/rados/operations/crush-map/](https://docs.ceph.com/en/latest/rados/operations/crush-map/) |
| Erasure coding | [https://docs.ceph.com/en/latest/rados/operations/erasure-code/](https://docs.ceph.com/en/latest/rados/operations/erasure-code/) |

Not sure how to lay out pools and device classes for your workload? ZSoftly designs the layout with you during [provisioning](https://docs.zcp.zsoftly.ca/cloud-storage/getting-started/provisioning#sizing-inputs).