Tiered storage allows different types of storage media to be used in a computing environment, with data stored in the different media based on the required access speeds and throughputs, with the foreknowledge that these different storage media also have different related costs. For example, storing data in-memory is generally always the most costly; storing the same data in remotely-attached Hard Disk Drives (HDDs) is going to be slower but also far less costly.
An example of different types of storage tiering currently popular:
In-memory — some data needs such high-speed, low-latency access that it is kept in-memory; usually for sub-millisecond speeds
Fast local NVMe SSD — locally-attached Solid State Drive storage; usually for single-digit millisecond speed access
Network-attached SSD — for example, Amazon Elastic Block Storage (EBS); network latency is added on top of the storage seek latency
Local or remote Hard Disk Drives (HDDs) — rotating media is slower both in terms of latencies and throughputs
Cloud object stores — here, the compute layer is decoupled from the storage layer, for example, Amazon S3, which, in itself, also has different storage tiers with access times that span from milliseconds to hours; the latter would be wholly unsuitable for real-time analytics but may have other use cases
Tiered storage solutions need not incorporate every possible combination of storage solutions. Specific options will be focused on as best fit for the use case and architectural goals.
For example, in-memory systems tend to be the most expensive and the fastest but are also the most volatile; data can be permanently lost. For use cases that require durability, local NVMe SSD is often chosen as the preferred “fast” storage medium.
To effectively use these different tiers, frequently-used data needs to be placed on the fastest media tier, and then less-frequently used data needs to be relegated to the slower tier media. The distributed system architecture needs to understand this design pattern and distribute data appropriate to the tier of storage it needs.
If you want to learn more about the benefits of Tiered Storage, take a look at some of the resources below:
Tim and Neha explain the need for a system that combines the speed of tightly-coupled systems and the cost-efficiency of decoupled systems.
Read the introduction of Tiered Storage and what improvement measurements we've been able to make using it.
See how the concept of Tiered Storage in StarTree Cloud compares to the wizarding world of Harry Potter in this lightboard video.