A study of the short message service of a nationwide cellular network

Petros Zerfos, Xiaoqiao Meng, Starsky H. Y. Wong, Vidyut Samanta, Songwu LuInternet Measurement Conference 2006 · DOI

TL;DR

This paper provides a comprehensive measurement study of Short Message Service (SMS) traffic on a nationwide cellular network in India. By analyzing over 59 million messages, the authors characterize message-level behavior, the "store-and-forward" mechanism, and conversational thread patterns, offering valuable insights into a ubiquitous yet often opaque communication service.

What Problem Does This Paper Address?

Despite the astronomical growth and widespread adoption of SMS as a primary communication tool for both personal and content-provider interactions, there's a significant lack of empirical understanding regarding its actual traffic patterns and user behavior in real-world nationwide cellular networks. Existing studies often rely on assumptions or simulations, failing to capture the nuances of actual SMS usage. This paper aims to bridge this gap by providing a data-driven analysis of SMS traffic, shedding light on how users communicate via SMS, how the service reliably delivers messages, and how conversations unfold.

Methods

The authors employed a measurement-based approach, collecting and analyzing Charging Data Records (CDRs) from a nationwide cellular carrier in India. They focused on three key areas:

  • Message-Level Characterization: Analyzing message size distribution, arrival patterns, and identifying the nature of senders and receivers (person-to-person vs. application-to-person).
  • Store-and-Forward Mechanism Analysis: Investigating message delivery times, the success rate of delivery, and the number of retries required for message delivery.
  • Thread-Level Characterization: Grouping messages into conversational threads based on sender-receiver pairs and time proximity, and analyzing thread duration and interarrival times.

Datasets

The primary dataset used in this study consists of anonymized SMS logs (CDRs) collected from a national cellular carrier in India over a period of three weeks. This dataset comprises over 59 million short messages exchanged by more than 10 million mobile users. The logs were collected from the billing subsystem and contain information such as timestamps, sender/receiver mobile IDs, message size, delivery status, and MSC association.

Key Insights and Ideas

  • Beyond Person-to-Person: A significant portion of SMS traffic isn't just between individuals. The study reveals that 7.2% of messages sent are requests to SMS services, and at least 10.1% of received messages originate from content providers, highlighting the growing role of application-driven communication.
  • SMS Reliability: While SMS is generally reliable (94.9% successful delivery), a non-trivial percentage of messages experience significant delays. A substantial portion of delivered messages take longer than a minute, and some even over an hour and a half, suggesting user-related factors (like phone being off) rather than just network latency.
  • Conversational Threads: SMS, though lacking explicit threading, exhibits conversational patterns. The majority of message exchanges between two users form "static threads" where users remain within the same Mobile Switching Center (MSC), simplifying mobility management for these interactions.
  • High-Volume Users: A small fraction of users send an exceptionally large number of messages, far exceeding typical human communication. This strongly suggests the involvement of automated services and content providers, further underscoring the shift towards application-driven SMS usage.

Key Results

  • Message Volume: Over 59 million short messages were exchanged by more than 10 million users during the three-week study period.
  • Delivery Success: 94.9% of messages were successfully delivered.
  • Delivery Latency:
    • 73.2% of delivered messages arrived within 10 seconds.
  • 17% took longer than one minute.
  • 5% required more than an hour and a half.
  • Thread Characteristics:
    • 87.1% of identified threads were "static" (users stayed within the same MSC).
  • The average thread duration was 8 minutes.
  • The average number of messages per thread was 4.9.
  • Application Traffic: 7.2% of total messages sent were requests to SMS services, and 10.1% of received messages were from content providers.

Limitations and Future Work

The authors acknowledge several limitations:

  • Single Carrier Data: The analysis is based on data from a single cellular carrier in India. Cross-validation with data from other locations is needed for broader generalization.
  • Thread Identification Assumption: The grouping of messages into threads relies on a timeout-based heuristic, which is somewhat arbitrary and cannot be definitively verified with the current logs.
  • Delivery Failure Reasons: While message delivery failures are categorized, the exact reasons (network, terminal, radio interface, general) could not be extracted from the CDRs.

Future work suggested includes:

  • Detailed Reliability Assessment: Systematically assessing SMS reliability, especially for mission-critical applications like emergency alerts.
  • Overload Incident Analysis: Investigating overload incidents, such as during New Year's Eve, to understand system behavior under stress.
  • Extracting Delivery Failure Reasons: Analyzing deeper traces from the SS7 protocol to pinpoint the causes of message delivery failures.
  • Characterizing Traffic at Thread/Session Level: Further refining the understanding of traffic patterns at the conversational thread or session level.
  • MMS Insights: Leveraging SMS traffic insights for planning and provisioning of future services like Multimedia Messaging Service (MMS).

Summary updated

BibTeX
@inproceedings{zerfos2006study,
  author = {Petros Zerfos and Xiaoqiao Meng and Starsky H. Y. Wong and Vidyut Samanta and Songwu Lu},
  title = {{A study of the short message service of a nationwide cellular network}},
  booktitle = {Internet Measurement Conference},
  year = {2006},
  doi = {10.1145/1177080.1177114},
}
Figure