Skip to content

Related papers

Jump to bottom
Puneet Mehrotra edited this page May 6, 2020 · 3 revisions

While reading related work for this project, I have uncovered some papers that I find important:

  • LiveGraph: A Transactional Graph Storage System withPurely Sequential Adjacency List Scans: I presented this paper at the NSS Seminar. The slides are available here

  • Do We Need Specialized Graph Databases? Benchmarking Real-Time Social Networking Applications: relevant, the focus of evaluation is real-time Social Network analysis on graph DB vs RDBMS. Ingest updates from a Kafka queue to model the real-time scenario. Verdict: no.

  • The case against specialized graph analytics engines This paper makes the claim that the reason that BSP/Pregel based Graph engines are so popular is because they present an easy programming paradigm to users. However, SQL is non-intuitive for expressing iterative/interactive graph processing algorithms. This paper presents GRAIL which is a DSL for expressing a graph analytics query in a SQL-ish language (?) that compiles down to optimised SQL that can run on standard issue RDBMS systems. They show that GRAIL is competitive with Giraph and GraphLab. They also make a really neat argument about scalability and dataset sizes: "Taking the number of iterations into consideration, the cluster runs 10+ times faster than a single node SQL Server. But 32 machines are used in that experiment. If we take into consideration all the costs such as hardware amortization cost, power cost, administration cost, SQL Server may be a better choice for graph analysis at this scale. Thus, while a lot of attention has focused on scale-out behavior of specialized graph engines, and the fact that papers report cluster numbers when the datasets are smaller than what can fit in typical single node today, we note that systems like RDBMSs that can deal with out-of-memory scenarios well are far more versatile and likely far more cost-effective than methods that demand an always-in-memory approach."

  • Graph Analysis – Do We Have to Reinvent the Wheel? Similar in idea to the Grail paper, but the focus of comparison is graphDBs to other RDBMS systems. They also present a DSL to make the interface for graph computation easier on indexed databases. Their perf is comparable.

Clone this wiki locally