Distributed graph processing

My main responsibility was to add fault-tolerance and elasticity to PGX.D, a distributed graph processing framework.

Initially, PGX.D has started as a prototype to evaluate the feasibility of a distributed graph processing framework. However, no fault-tolerance or elasticity was built into the system.

My work has been to add these features to the system, which included integrating the Raft consensus protocol as well as adding a level of indirection for all graph-related data in order to facilitate dynamic graph partitioning and recovery in case of hardware failures.

In the process of developing these features, I have also been building a distributed health monitoring system for PGX.D.

Patents:

• Fast distributed graph query engine
• Language interoperable runtime adaptable data collections

Interns:

• Lukas Werling

Runtime optimizations

Following up on my work around "Shoal" at ETH, I have been helping to integrate NUMA optimizations in our shared C++/Java runtime used in PGX.

We have been using the GraalVM to efficiently share low-level code between the C++ and Java runtime.

Publications and patents:

• Paper: Analytics with Smart Arrays: Adaptive and Efficient Language-Independent Data
• Patent: Language interoperable runtime adaptable data collections

Misc

During my time at Oracle, I also had the honor to host a "Rust meetup". My slides are here:

• Presentation
• Rust project
• Speed comparison Java + Python