Revisiting next-hop selection in multipath networks

Abstract

Multipath routing strategies such as Equal-Cost MultiPath (ECMP) are widely used in IP and data-center networks. Most current methods to balance packets over the multiple next hops toward the destination base their decision on a hash computed over selected fields of the packet headers. Because of the non-invertible nature of hash functions, it is hard to determine the values of those fields so as to make the packet follow a specific path in the network. However, several applications might benefit from being able to choose such a path. Therefore, we propose a novel next-hop selection method based on an invertible function. By encoding the selection of successive routers into common fields of packet headers, the proposed method enables end hosts to force their packets to follow a specific path.

FatPaths: Routing in Supercomputers and Data Centers when Shortest Paths Fall Short

“…Multi-pathing Many works on multi-pathing exist [4,20,20,26,27,27,28,53,65,76,91,102,109,127,131,134,134,135,144,155]. Our work differs from them all: it focuses on path diversity in lowdiameter topologies, it considers both minimal and non-minimal paths, and it shows a routing scheme using the explored path diversity.…”

On Analysis of Traffic Flow Demultiplexing Effectiveness

“…В измененном варианте предотвращается маршрутизация родственных подпотоков через один и тот же следующий скачок, если это возможно. Реализовать предложенный подход в сетях ISP можно при помощи стратегии, предложенной в [20], которую будем называть Обратимой Функцией Хэширования (ОФХ) или Invertible Hash Function (IHF). Идея ОФХ заключена в том, чтобы выбор среди множества альтернативных следующих скачков осуществлялся при помощи обратимой хэш-функции от заголовков пакетов.…”

Dynamic load distribution with hop-by-hop forwarding based on max-min one-way delay

“…Note that node b is a candidate too, so node a adds b to ANS(a, d). Then node a chooses the optimal next node according to (9).…”

“…As we can see, owing to the fact that DDMM is a little more complicated than FLARE, the e2e throughput of DDMM is lower than FLARE, but the difference is very small and almost can be ignored. where the X axis is the node's number and the Y axis is the number of packets, shows the gap of the number of received packets between each pair of nodes (8,9), (10,11) and (12,13). As can be observed, PWFR, under the routing weight vector of (0.5, 0.5), can control the amount of the load assigned to each node.…”

“…The dynamic load distribution algorithm is based on the dynamic changes of network traffic and network status that cannot be predicted in advance, by assigning load to multiple links, multiple servers, etc., to achieve load balancing. Adaptive load distribution can effectively improve the performance and scalability of the network, allow maximizing network throughput and optimize resource utilization [9].…”