Experiment coordination for large-scale measurement platforms

Mario A. Sánchez*, Fabián E. Bustamante*, Balachander Krishnamurthy‡, Walter Willinger†
(*) Northwestern University (‡) AT&T Research Lab (†) Niksun, Inc.


The risk of placing an undesired load on networks and networked services through probes originating from measurement platforms has always been present. While several scheduling schemes have been proposed to avoid undue loads or DDoS-like effects from uncontrolled experiments, the motivation scenarios for such schemes have generally been considered “sufficiently unlikely” and safely ignored by most existing measurement platforms. We argue that the growth of large, crowdsourced measurement systems means we cannot ignore this risk any longer.

In this paper we expand on our original lease-based coordination scheme designed for measurement platforms that embrace crowdsourcing as their method-of-choice. We compare it with two alternative strategies currently implemented by some of the existing crowdsourced measurement platforms: centralized rate-limiting and individual rate limiting. Our preliminary results show that our solution outperforms these two naive strategies for coordination according to at least two different intuitive metrics: resource utilization and bound compliance. We find that our scheme efficiently allows the scalable and effective coordination of measurements among potentially thousands of hosts while providing individual clients with enough flexibility to act on their own.