[Paper Review] Distributed Publish/Subscribe Query Processing on the Spatio-Textual Data Stream

Zhida Chen, Gao Cong+3 more|arXiv (Cornell University)|Dec 8, 2016

Peer-to-Peer Network Technologies19 references3 citations

TL;DR

This paper proposes PS² Stream, a distributed publish/subscribe system for real-time processing of spatio-textual data streams, using a hybrid workload partitioning strategy that jointly optimizes spatial and textual properties to minimize total workload and balance worker loads. The system achieves 26% higher throughput with low migration cost through dynamic load adjustment, outperforming baselines in latency and scalability on real-world Twitter data.

ABSTRACT

Huge amount of data with both space and text information, e.g., geo-tagged tweets, is flooding on the Internet. Such spatio-textual data stream contains valuable information for millions of users with various interests on different keywords and locations. Publish/subscribe systems enable efficient and effective information distribution by allowing users to register continuous queries with both spatial and textual constraints. However, the explosive growth of data scale and user base has posed challenges to the existing centralized publish/subscribe systems for spatio-textual data streams. In this paper, we propose our distributed publish/subscribe system, called PS2Stream, which digests a massive spatio-textual data stream and directs the stream to target users with registered interests. Compared with existing systems, PS2Stream achieves a better workload distribution in terms of both minimizing the total amount of workload and balancing the load of workers. To achieve this, we propose a new workload distribution algorithm considering both space and text properties of the data. Additionally, PS2Stream supports dynamic load adjustments to adapt to the change of the workload, which makes PS2Stream adaptive. Extensive empirical evaluation, on commercial cloud computing platform with real data, validates the superiority of our system design and advantages of our techniques on system performance improvement.

Motivation & Objective

• To address the scalability limitations of centralized publish/subscribe systems under growing spatio-textual data and user workloads.
• To design a distributed system that efficiently partitions both data and query workloads across workers while minimizing total processing load.
• To achieve load balancing and low migration cost in response to dynamic workload changes in real time.
• To evaluate the system's performance on real-world spatio-textual data with high-throughput and low-latency requirements.

Proposed method

• Proposes a hybrid workload partitioning algorithm that jointly considers spatial regions and textual keyword patterns to distribute data and query streams effectively.
• Introduces a workload distribution model that minimizes total workload and balances worker utilization by analyzing spatial and textual correlations.
• Designs dynamic load adjustment mechanisms—local and global—that reassign workloads with minimal data migration and low latency impact.
• Employs a cell-based data partitioning scheme where each worker manages a spatial-textual cell, enabling efficient workload tracking and migration.
• Uses real Twitter data streams with timestamps to scale workloads in experiments, simulating dynamic query distributions over time.
• Evaluates system performance using throughput, latency, migration cost, and migration time metrics on Amazon EC2.

Experimental results

Research questions

• RQ1 How can workload be optimally partitioned across distributed workers when handling both spatio-textual data and subscription queries?
• RQ2 What partitioning strategy minimizes total system workload while maintaining load balance in a distributed publish/subscribe system?
• RQ3 How can dynamic load adjustments be performed with low migration cost and minimal impact on processing latency?
• RQ4 To what extent does dynamic load adjustment improve system throughput under changing workloads?

Key findings

• PS² Stream achieves 26% higher throughput compared to a system without dynamic load adjustments, demonstrating the effectiveness of its load adjustment mechanism.
• The GR (Global Rebalancing) algorithm incurs the lowest migration cost and time, reducing migration time by nearly 50% compared to RA and 70% compared to SI.
• GR achieves the best latency profile, with 80% of tuples unaffected by migration (under 100ms), and only 4% significantly delayed (over 1 second).
• DP (Dynamic Partitioning) performs second-best in latency, with 72% of tuples unaffected, but incurs higher selection time due to cell selection overhead.
• The hybrid workload partitioning strategy considering both space and text outperforms baselines that use only spatial or only textual partitioning in throughput and load balancing.
• Migration cost and time increase with higher query volume, but GR and DP maintain significantly lower overhead than SI and RA under all workloads.