5 Partitioning
This chapter covers
- Understanding the benefits and downsides of partitioning
- Partitioning strategies and how to choose one
- Request routing when partitioning
- Skewed workloads and hot partitions and how to mitigate against them
In the previous two chapters, we explored techniques for reducing latency with colocation and replication. Colocation places related computing, such as business logic and data resources, nearby, which minimizes the network distance between them. Colocation can improve latency by reducing communication latency between components. For example, an application that uses serverless functions for backend logic may benefit from a database that is colocated with the serverless runtime. Replication, on the other hand, is a technique for copying relevant data at multiple locations while maintaining consistency between the copies. With replication, you receive the same benefits as with colocation, but across various locations. However, having the entire data set replicated across numerous locations can be impractical due to storage costs and network bandwidth requirements. Additionally, maintaining consistency across multiple replicas can introduce significant coordination overhead, as every write operation may need to be synchronized across various locations, thereby creating latency bottlenecks rather than reducing them.