Mitigating transmission control protocol Incast congestion collapse with adaptive resolution round trip time measurement
Lawrence A. Stewart · Swinburne figshare (Swinburne University of Technology) · 2024
Data-driven services delivered while the consumer waits are intrinsically sensitive to service delays impacting consumer experience. This can transfer to the financial bottom line when commercial interests are involved, such as with Internet web search and e-commerce. This thesis grows the body of knowledge regarding Incast congestion, a source of significant variable delay in delivering such services that rely on distributed computation. By developing free tools and using them to analyse the Incast phenomenon, a novel mitigation algorithm was developed which can improve service delivery times for Incast affected transactions, sometimes by more than an order of magnitude.