Title: Putting consistency back into eventual consistency
By: Valter Balegas (NOVA LINCS / FCT-UNL)
Abstract:
Geo-replicated storage systems are at the core of current Internet services. The designers of the replication protocols used by these systems must choose between either supporting low-latency, eventually-consistent operations, or ensuring strong consistency to ease application correctness. We propose an alternative consistency model, called Explicit Consistency, that strengthens eventual consistency with a guarantee to preserve specific invariants defined by the applications. Given these application-specific invariants, a system that supports Explicit Consistency identifies which operations would be unsafe under concurrent execution, and allows programmers to select either violation-avoidance or invariant-repair techniques.
In the first part of this talk we present Indigo, a system that implements Explicit Consistency. Indigo guarantees strong application invariants while providing similar latency to an eventually-consistent system in the common case, by making good use of violation-avoidance and simple invariant-repair techniques. The shortcoming of the approach is that some operations may become unavailable for execution when replicas become unavailable. In the second part of the talk we discuss how to overcome these limitations. The idea is to improve the applicability of invariant-repair techniques, which are more suitable to the geo-replicated environment since they do not require ensuring safety before operation execution.
Our ultimate goal is to improve the quality of service of services that operate at planetary-scale, by making them highly available without impairing correctness.