Minimum-Delay Multicast Algorithms for Mesh Overlays

Abstract

We study delivering delay-sensitive data to a group of receivers with minimum latency. This latency consists of the time that the data spends in overlay links as well as the delay incurred at each overlay node, which has to send out a piece of data several times over a finite-capacity network connection. The latter part is a significant portion of the total delay as we show in the paper, yet it is often ignored or only partially addressed by previous multicast algorithms. We analyze the actual delay in multicast trees and consider building trees with minimum-average and minimum-maximum delay. We show the NP-hardness of these problems and prove that they cannot be approximated in polynomial time to within any reasonable approximation ratio. We then present a set of algorithms to build minimum-delay multicast trees that cover a wide range of application requirements-min-average and min-max delay, for different scales, real-time requirements, and session characteristics. We conduct comprehensive experiments on different real-world datasets, using various overlay network models. The results confirm that our algorithms can achieve much lower delays (up to 60% less) and up to orders-of-magnitude faster running times (i.e., supporting larger scales) than previous related approaches.

Overlay Multicast Update Strategy Based on Perturbation Theory

“…The calculation of the multicast fluctuation factor's complexity is mainly divided into two parts: First is to measure the performance of all the multicast nodes and the performance of the link between nodes when the cost matrix is obtained. The complexity degree is It can be known from the analysis of the complexity of traditional multicast algorithm that, [15], and the maximum calculation complexity of SOM-HS is ( log ) O nk n (in which k is the node out degree) [21]. From this it can be seen that even if the multicast metric matrix is a sparse triangular matrix, the Musp algorithm is more complex than traditional algorithm.…”

“…In order to verify the validity of the multicast fluctuation factor, this paper chooses three multicast algorithms to analyze the relationship between multicast fluctuation factor and other three factors: data retransmission times, maximum delay and delay variation. The three kinds of overlay multicast algorithms are NICE [15], SMO [12] and SOM-HS [21] respectively. The multicast's basic idea of the three algorithms is the following respectively: NICE algorithm constructs multicast tree through the stratification clustering method.…”

“…Local update strategy means when links with the following characteristics like heavy congestion [8][9], higher bandwidth [10], more stable parent nodes [11][12] and a closer network distance [13][14] are detected, we will start the local updating of the multicast tree, but the local update cannot guarantee a global optimal multicast. Global update means that the overall performance of the multicast tree is unable to meet the needs of business applications, for example the maximum delay exceeds the threshold [15] and the data retransmission rate is high [16] and so on. At this time, the entire multicast tree needs to be reconstructed.…”

BDS+: An Inter-Datacenter Data Replication System With Dynamic Bandwidth Separation

“…Note that BDS+ selects application-level overlay paths, and is therefore complementary to network-layer optimization of WAN performance. While application-level multicast overlays have been applied in other contexts (e.g., [9], [10], [11], [12]), building one for inter-DC multicast traffic poses two challenges. First, as each DC has tens of thousands of servers, the resulting large number of possible overlay paths makes it unwieldy to update overlay routing decisions at scale in real time.…”