A09: Ring: Unifying Replication and Erasure Coding to Rule Resilience in KV-Stores
Student: Konstantin Taranov (ETH Zurich)
Supervisor: Torsten Hoefler (ETH Zurich)
Abstract: There is a wide range of storage schemes employed by KV-stores to ensure reliability of stored keys. However, previous implementations do not allow choosing the storage scheme dynamically, thereby forcing developers to commit to a single scheme. Such inefficient data management wastes cluster resources such as memory usage, network load, latency, availability and many others.
To solve this problem, we have designed a strongly consistent key-value store Ring that empowers its users to explicitly manage storage parameters like other resources, such as memory or processor time. The key feature of Ring is that all keys live in the same strongly consistent namespace and a user does not need to specify the resilience when looking up a key or value.
Our poster demonstrates how future applications that manage resilience of key-value pairs consciously can reduce the overall operational cost and improve performance significantly.
ACM-SRC Semi-Finalist: no
Poster: pdf
Two-page extended abstract: pdf
Poster Index