We are bringing a number of improvements to GraphScope, alongside the GraphScope 0.16.0 release. This release introduces many new features on backend engines and system stability. We completely remove a legacy Graph Interactive Engine (GIE), while officially replacing it with the latest version based on an intermediate representation (IR) layer. The Graph Learning Engine (GLE) now supports real-time sampling on dynamic graphs. In addition, the Jave SDK of Graph Analytics Engine (GAE) can work on MacOS. Meanwhile, we start to release a nightly version every day, and you can try it with
pip3 install graphscope --pre.
We highlight the following improvements included in this release:
- Enhancements of the GIE GAIA-IR engine:
- Completely replacing the legacy GIE engine with the latest version based on IR layer;
- Optimizing the logic of lazily fetching properties from the graph store, and implementing a new CSR-based storage for experiment purpose;
- New language features including more grouping functions, more options for group/dedup/orderby keys, etc.
- Dynamic-Graph-Service (DGS) in GLE:
- DGS is an online inference service, and it supports real-time graph sampling for GNN algorithms on dynamic graphs with streaming graph updates;
- It provides a performance guarantee of sampling P99 latency in 20ms on large-scale dynamic graphs.
- Other enhancements:
- Enabling Jave SDK of Graph Analytics Engine (GAE) on MacOS;
- A series of enhancements on the persistent graph storage
For more detailed improvements that have been made in this release, please refer to the complete changelog.