A recently proposed congestion control algorithm (CCA) called BBR (Bottleneck Bandwidth and Round-trip propagation time) has shown a lot of promise in avoiding some of the problems that have plagued loss-based CCAs. Nevertheless, deployment of a new alternative algorithm requires a thorough evaluation of the effect of the proposed alternative on established transport protocols like TCP CUBIC. Furthermore, evaluations that consider the heterogeneity of Internet trafﬁc sizes would provide a useful insight into the deployability of an algorithm that introduces sweeping changes across multiple algorithm components. Yet, most evaluations of BBR’s impact and competitive fairness have focused on the steady-state performance of large ﬂows. This work expands on previous studies of BBR by evaluating BBR’s impact when the trafﬁc consists of ﬂows of different sizes. Our experiments show that under certain circumstances BBR’s startup phase can result in a signiﬁcant reduction of the throughput of competing large CUBIC ﬂows and the utilization of the bottleneck link. In addition, the steady-state operation of BBR can have negative impact on the performance of bursty ﬂows using loss-based CCAs over bottlenecks with buffer sizes as high as two times the bandwidth-delay product.
|Status||Publicerad - 2018|
|Evenemang||2018 30th International Teletraffic Congress - |
Varaktighet: 1980-jan.-01 → …
|Konferens||2018 30th International Teletraffic Congress|
|Period||80-01-01 → …|
- Datavetenskap (10201)