"If you’re concerned about scalability, any algorithm that forces you to run agreement will eventually become your bottleneck." Failure tolerant distributed key-value stores that provide strong consistency need to run an agreement protocol among a set of nodes. Such a protocol either involves much data to be sent among these nodes, in terms of messages, or high latency due to the number of communication steps between these nodes. Paxos is an example for a fault tolerant consensus algorithm used in a number of distributed key-value stores.
The presentation explains the theoretical background about Werner Vogels's claims about distributed key-value stores with strong consistency. It covers the CAP theorem as well as Paxos as an example for an agreement protocol of a fully decentralized key-value store.
Speaker's bio: Monika Moser is currently working as a Erlang/Ruby/Hadoop developer at nugg.ad. Before starting at nugg.ad she designed and implemented the transactional storage layer of Scalaris, a scalable, transactional, distributed key-value store written in Erlang, while working in research at Zuse Institut, Berlin following her graduation from Humboldt Universität.