UlTraMan: a unified platform for big trajectory data management and analytics

[Sunt-ing]

link: http://www.vldb.org/pvldb/vol11/p787-ding.pdf

[Sunt-ing]

What problem does the paper solve? Is it important?

Problem:

• Although there are many methods designed for specific trajectory data management and analytics needs, there is no holistic geospatial system in the big data scenario.
• Original Spark is not suitable for random data access, which is very normal in trajectory data management.

Importance:

• Yes.
• Trajectory data analysis is not a need as comprehensive as OLAP or graph, but there are still many scenarios for it.
• There is no holistic big data system for trajectory data analysis before this paper.

How does it solve the problem?

• Builds a system that supports and optimizes many usual queries on Spark, and provides an HTTP interface besides original Spark interface.
• Integrating a KV store into Spark to enable random data access.

How does this work relate to other research?

• PIST, BerlinMOD, TrajStore, SharkDB: designed for centralized architecture and thus have limited flexibility.
• SpatialHadoop, Simba: unable to exploit the characteristics of trajectory data efficiently.
• CloST, PARADASE, Elite: use specific partitioning strategies and fail to be a holistic solution that supports customizable data formats, partitioning strategies, index structures, processing methods, and analysis techniques.
• SnappyData, Apache Ignite, IndexedRDD: only enhance Spark and eliminate inefficiencies of heterogenous systems, but no operations and optimizations for trajectory data analytics.

What could be improved?

• Support streaming data updates and queries.

Others

• To show that it's well-pluggable, it shows the system layer by layer.
• Aalborg University is known for geospatial research: Yufei Tao, Ken(ML Yiu), Christian Jensen, Lu Chen... so many heard names.
• Systematic contribution will make the DB work more solid. The core part of the systematic contribution can be as small as adding an existing KV to Spark, but the contribution can be amplified by adding complete performance research && key improvement, as well as some downstream design derived from the core.