Geo-distribution of actor-based services

Abstract

Many service applications use actors as a programming model for the middle tier, to simplify synchronization, fault-tolerance, and scalability. However, efficient operation of such actors in multiple, geographically distant datacenters is challenging, due to the very high communication latency. Caching and replication are essential to hide latency and exploit locality; but it is not a priori clear how to combine these techniques with the actor programming model.We present Geo, an open-source geo-distributed actor system that improves performance by caching actor states in one or more datacenters, yet guarantees the existence of a single latest version by virtue of a distributed cache coherence protocol. Geo's programming model supports both volatile and persistent actors, and supports updates with a choice of linearizable and eventual consistency. Our evaluation on several workloads shows substantial performance benefits, and confirms the advantage of supporting both replicated and single-instance coherence protocols as configuration choices. For example, replication can provide fast, always-available reads and updates globally, while batching of linearizable storage accesses at a single location can boost the throughput of an order processing workload by 7x.

CScript: A distributed programming language for building mixed-consistency applications

“…Geo [29] is an actor system for geo-replication that combines caching with replication techniques to hide latency and benefit from data locality where possible. Geo supports "single-instance" and "multi-instance" caching policies for actors across clusters.…”

A fault-tolerant programming model for distributed interactive applications

“…None of these systems is applicable without central coordination and replacements for failed devices, which makes sense in their environment, but not for interactive applications. Bernstein et al [2017] introduce an eventually consistent mechanism to synchronize state between multiple data centers, however, the solution does not extend to client devices. GSP [Burckhardt et al 2015] provides a semantic foundation for eventual consistency using a central server connected to clients with unreliable connections; GOS [Gotsman and Burckhardt 2017] generalizes GSP and provides a formal foundation for proving equivalence of multiple implementations.…”

“…Existing research shows that many common data types can be expressed as CRDTs [Shapiro et al 2011a]. Their wide-spread usage confirms the practicality of CRDTs, to replicate data over unreliable connections [Bernstein et al 2017;Shapiro et al 2018].…”

Reactive caching for composed services: polling at the speed of push

“…Actor Frameworks. Actor frameworks [Akka 2016;Armstrong 2010;Bernstein et al 2017;Chuang et al 2013;Orbit 2016;Orleans 2016;Sang et al 2016; SF Reliable Actors 2016] directly embody the composed-services paradigm for building scalable distributed systems. Actors are not only suitable for formal study, but can also deliver excellent performance in practice, and distribute easily over elastic clusters since they do not share memory and communicate asynchronously.…”