The Early Original Theory of the SICS Team Successfully Landed in the Industry,Contribute Power to the International Open-Source Community

CategoryNews 394

With the rapid development of the digital economy, graph computing is prevalent in the precision marketing of social networks, financial anti-fraud, and e-commerce. Recently, at the 2nd World Science and Technology Development Forum, Alibaba officially released GraphScope, a one-stop ultra-large-scale distributed graph computing platform, which will be open-source in December. It worth be noted that GAE (Graph Analytics Engine), one of the key components that support GraphScope to process trillion-scale graph data, originated from the graph computing engine GRAPE developed by the chief scientist of SICS.

The GRAPE engine was first proposed in “Parallelizing Sequential Graph Computations” published by the research team led by the chief scientist of SICS. The GRAPE engine is an original large-scale graph data distributed graph computing platform. It supports the plug-and-play stand-alone algorithm. Ordinary users can complete the parallelization of the algorithm on the GRAPE system by providing simple serial graph algorithms and a few parameters, thus free users from barriers to using parallel computing of big data. GRAPE makes it possible to universalize graph data computing applications.

Figure 1: GAE is originated from GRAPE developed by the chief scientist of SICS(Source:

In addition to its leading accessibility, the computing speed of GRAPE is also far ahead. Relying on original parallel algorithms and parallel models, it can calculate 5.22 billion times per second, enabling massive graph data processing in a short time. Proven by the industry’s authoritative LDBC benchmark test, GRAPE has 28 test data sets for six types of algorithms, 26 of which achieve second-level operations. The overall performance is nearly 52 times higher than others in the industry. What’s more interesting is that using LCC (local clustering coefficient) on GRAPE to analyze a social network composed of 56 million users and their 1.8 billion friend relationships takes only about 1 minute to calculate the probability that any two of them are friends. The computing resources only require four sets of Alibaba Cloud memory-type ECS (Elastic Compute Service). Its computing power consumption and running speed can be guaranteed synchronously, while many similar graph computing products still cannot support the LCC algorithm.

The powerful computing performance of GRAPE also gains recognization from the international academic community. Related papers have successively won the best paper award of SIGMOD 2017, the best presentation award of VLDB 2017, and the research hotspot award of SIGMOD 2018.

Though GRAPE’s technology application on GraphScope is just a casual demonstration in the industry of the early research results led by the chief scientist of SICS, it further verifies the industrial productivity behind the big data system and theory innovation.

The SICS research teams will keep concentrate on computing science systems and theories represented by big data, developing core system software, and creating an innovative model of S (Science) T (Technology) E (Engineering). As a result, SICS will form a closed loop of a completely self-developed industry from theoretical breakthroughs, technical innovation, and industrialization.