Tiered storage divides storage devices into a cache tier and a capacity tier according to the characteristics of different storage media in the cluster, balancing performance and capacity. When there are different types of physical disk combinations in the cluster, such as NVMe SSD + SATA SSD or NVMe SSD + HDD, or if the cluster requires the enabling of certain space optimization features (such as erasure coding), the cluster should adopt a tiered storage mode. In this mode, ABS transforms most of the space on the higher-performance physical drives into a cache layer. The cache layer is further divided into write cache (a unified cluster-wide cache layer) and read cache (disk-local read acceleration for the capacity tier). The lower-performance capacity-type physical drives are pooled into the capacity tier.
When a volume is written to, the system typically writes data to the cache tier first. If the logical data area of the volume is frequently updated, the data will remain in the cache tier for a longer period of time. Conversely, data that is not frequently updated periodically sinks to the capacity tier. Data that is infrequently updated but frequently accessed is stored in the read cache of the capacity tier to ensure high-speed access.
Data in the cache tier is stored using replications, while the organization of data in the capacity tier is determined by the storage policy set by the user for the volume. In the tiered storage mode, for critical volumes that require performance assurance, ABS also supports configuring the volume pinning mode. In this mode, all data of a volume is retained in the cache tier, regardless of whether it is hot or cold. This meets the application scenarios where certain data always requires high-performance, low-latency access.
