Giza: Erasure Coding Objects across Global Data Centers

Microsoft Azure Storage is a global cloud storage system with a footprint in 38 geographic regions. To protect customer data against catastrophic data center failures, it optionally replicates data to secondary DCs hundreds of miles away. Using Microsoft OneDrive as an example, this paper illustrates the characteristics of typical cloud storage workloads and the opportunity to lower storage cost for geo-redundancy with erasure coding.

The paper presents the design, implementation and evaluation of Giza – a strongly consistent, versioned object store that applies erasure coding across global data centers. The key technical challenge Giza addresses is to achieve single cross-DC round trip latency for the common contention-free workload, while also maintaining strong consistency when there are conflicting access. Giza addresses the challenge with a novel implementation of well-known distributed consensus algorithms tailored for restricted cloud storage APIs. Giza is deployed to 11 DCs across 3 continents and experimental results demonstrate that it achieves our design goals.