CloudBDT – Algorithms for large data streams in cloud networks
A fast connection does not automatically make a long-distance transfer fast. Large amounts of data can be in flight at once, while losses and growing queues make delivery harder. CloudBDT investigated algorithms and methods for reliable transport of large datasets in cloud environments.
The work connected several layers: RMDT transport to multiple receivers, delay-based congestion control, available-bandwidth estimation and efficient packet processing at the endpoints. The recurring question was how to increase useful delivery while understanding the additional delay created along the way.
© FILA
BQL congestion control provides a concrete example. It uses delay information to estimate queue occupancy at the bottleneck and adjust the sending rate. Published experiments with a Netropy 40G WAN emulator compared BQL and TCP BBR under defined bandwidth, delay and loss conditions. The controlled conditions make the relationship between useful delivery and delay traceable.
CloudBDT forms part of the scientific background to current FILA work on transport control and data distribution. RMDT explains the protocol approach; AI-RMDT investigates its development using learning methods.
Project and funding
Status: completed. Funding: Volkswagen Foundation, through a trilateral collaboration of researchers from Germany, Ukraine and Russia.
BQL: setup and results · Funding acknowledgement in the ICAIIT proceedings
