How We Built a Geo-Distributed Database With Low Latency

In modern cloud architectures, the holy grail is to be able to operate across multiple geographic regions concurrently, and tolerate the failure of either an availability zone or an entire region. This is a challenging problem for databases​, which typically requires a tradeoff, such as higher latency for client connections, or a weaker level of consistency. For example, in MySQL​ it is common to use asynchronous replication in order to keep latency low.

In this talk, Ming Zhang will cover TiDB’s approach to geographically distributing data. It includes features such as the ability to schedule data placement to different locations in order to provide low latency, as well as ‘stale reads’ which can read a local copy of data that may be slightly delayed from the primary copy.

Speaker: Ming Zhang – PingCAP