Uber Technologies, Inc.
Table replica allocation in a replicated storage system

Abstract:

A replicated storage system allocates tables to storage clusters to balance the usage of resources in the replicated storage system. The replicated storage system determines the query load of the stored tables and generates candidate table allocations based on the query loads. The candidate table allocations are generated based on table allocation criteria based on the popularity of the tables or whether the tables are related to other tables. The replicated storage system determines resource parameters for each candidate table allocation and generates a resource usage score for each candidate table allocation based on a utility function that is weighted based on priorities of the resources used by the replicated storage system. The replicated storage system selects a table allocation from the candidate table allocations based on the resource usage scores of the candidate table allocations and allocates table replicas to the storage clusters.