Meddle: middleboxes for increased transparency and control of mobile traffic

Ashwin Rao, Justine Sherry, Arnaud Legout, Arvind Krishnamurthy, Walid Dabbous, David R. ChoffnesStudentworkshop@conext 2012 · DOI

TL;DR

Mobile networks are a black box for users and researchers, rife with privacy issues and opaque traffic manipulation. This paper introduces Meddle, a framework that combines VPNs and middleboxes to provide unprecedented transparency and control over mobile traffic, enabling new avenues for research and user empowerment.

What Problem Does This Paper Address?

The mobile internet ecosystem is notoriously difficult to understand and control. Users are often at the mercy of closed-source apps and network providers who can transparently modify, block, or interfere with their traffic. For researchers, characterizing this traffic and experimenting with new network services is a significant challenge, often requiring intrusive methods like jailbreaking devices, which can still be circumvented by network-level manipulations. This lack of transparency hinders both user privacy and the development of better mobile network services.

Methods

Meddle tackles this problem by creating a novel framework that leverages two existing technologies: Virtual Private Networks (VPNs) and middleboxes.

  • VPNs for Interception: Meddle uses VPN tunnels to route all mobile device traffic through a designated Meddle server. This approach is device-agnostic (works across Android, BlackBerry, iOS) and bypasses the need for jailbreaking, as native VPN support is common. This provides a continuous and comprehensive view of mobile traffic, regardless of the underlying wireless technology or carrier.
  • Middleboxes for Control and Analysis: Once traffic reaches the Meddle server, it can be processed by custom "middlebox" services. These services can perform actions like packet filtering (e.g., for ad blocking), caching, intrusion detection, and more. This allows for the implementation of network-level services that benefit users and enable researchers to experiment with new network functionalities as if they were deployed in carrier networks.

The core innovation lies in combining these two technologies in an "unintended" way to create an experimental platform that aligns user and researcher interests.

Datasets

Not discussed in the paper. The paper mentions a prototype that has been running since August 27, 2012, serving mobile devices including Android phones, iPhones, iPads, and iPod Touches. Participants in their study are using their own devices.

Key Insights and Ideas

  • Bridging the Research-User Divide: Meddle offers a unique opportunity to create a positive feedback loop where practical research artifacts (like ad blockers) incentivize user adoption, which in turn provides more data for further research.
  • Unlocking Network-Level Control for Users: By providing a platform for middlebox deployment, Meddle empowers users with control over their traffic that was previously only available to network providers.
  • Separating Network Performance: Meddle allows for the separation of mobile network performance from server-side performance, which is crucial for accurate bottleneck identification in mobile applications.
  • Simulating Carrier-Level Experiments: Researchers can use Meddle to test the impact of new middleboxes (e.g., anonymization systems) in a controlled environment, mimicking carrier-level deployments without needing actual carrier cooperation.

Key Results

  • Ad Blocking Efficacy: An implemented DNS-based ad filter reduced total traffic by 0.05% to 0.8% on mobile devices. The authors are investigating why this reduction might differ from previous work.
  • Traffic Volume Analysis: A study summarized in Table 1 showed that an iPhone consumed more cellular bandwidth than an Android phone, and that the vast majority of mobile traffic was traversing Wi-Fi networks.
  • Overhead: Tunneling data traffic via VPN resulted in low overheads. Power consumption increased by about 10% when streaming HD video. Data overhead from IPsec headers and keep-alive messages ranged from 8-12% for Android phones and iPhones.
  • Latency: Latencies observed from PlanetLab nodes were between 3 ms and 13 ms (median 5 ms). The authors argue that this additional latency is expected to be relatively small or negligible compared to typical mobile network latencies.

Limitations and Future Work

  • Limitations: The paper acknowledges that the ad blocking filter might be incomplete, and user-specific behavior could influence observed traffic reductions. The overhead for data traffic is noted as 8-12%.
  • Future Work:
    • Recruiting more users for an IRB-approved study on mobile network usage profiles.
  • Allowing users to observe how their apps use the network and share data (similar to Mozilla Collusion).
  • Testing algorithms for content coalescing, caching, prefetching, and DOM processing offloading.
  • Making the Meddle system publicly available and developing a framework for researchers to contribute and analyze data.

Summary updated

BibTeX
@inproceedings{rao2012meddle,
  author = {Ashwin Rao and Justine Sherry and Arnaud Legout and Arvind Krishnamurthy and Walid Dabbous and David R. Choffnes},
  title = {{Meddle: middleboxes for increased transparency and control of mobile traffic}},
  booktitle = {Studentworkshop@conext},
  year = {2012},
  doi = {10.1145/2413247.2413286},
}
Figure