First, let’s explain what distributed and cloud native are. The concept of distribution is relatively easy to understand, which means that various components of the system collaborate and divide work on different nodes to achieve larger-scale and higher-performance processing. Cloud native does not only refer to services running on Kubernetes (K8s) or on public clouds. In my understanding, distributed cloud native means that the system is first and foremost a resource pool, including not only computing resources, but also data and other resources.
For example, vector retrieval usually requires a lot of computing resources during the index building phase, but once the index construction is completed, these computing resources are wasted. In traditional approaches, additional management and optimization may be required to avoid waste. However, in a distributed native system, resources can be shared among multiple tenants, thereby improving resource utilization. This also means that for a given user, the index building process can be accelerated through a shared pool of resources.
This is the concept of resource pool, which allows multiple tenants to share the same computing resource pool and improve resource reuse rate. From a user's perspective, with the same resource investment or the same cost, by utilizing this resource pool, more resources can be allocated, thereby speeding up index construction.
This can improve efficiency while maintaining the same resource expenditure, because the value of resources can be measured by the usage per unit time. Through this elastic allocation of resource pools, users can obtain more resources at the same cost, thereby shortening task execution time and improving user experience.