The distribution of large data sets from a centralized node to several destination sites is frequently required by many data-intensive networking applications; this distribution can be efficiently achieved by the means of multicasting. Multicasting has been typically considered for on-demand applications and services, e.g., video-on-demand, IPTV, etc., which usually require start of data transmission immediately. We consider that multicast sessions can be provisioned starting with flexible times and that the multicast client can specify a maximum allowed time by which all data needs to be delivered to destinations. This is true for e-Science and high-performance applications, in which data distribution is not necessarily immediate. In this paper, we study the problem of provisioning dynamic multicast data-distribution requests (MDDRs) with flexible scheduling over optical WDM networks. We consider the practical case of fractional-capacity multicast sessions that require less than the entire wavelength capacity (nodes are equipped with multicast-capable opaque switches). We devise provisioning methods based on the multicast tree (or light-tree) distribution model. In our first approach (named Rand), we generate multiple randomized alternate trees on which we try to provision the multicast session and then assign wavelengths and schedule the session. In our second approach (named AllSlots), we dynamically generate light-trees depending on the network state. In our next approach (named Break), when provisioning an entire tree fails, we try to “break” the tree into time-independent subtrees. We also study the impact of allowing data to be buffered at intermediary nodes and then transmitted toward destinations (method named Buffer) and consider an approach that partitions the data sets. Finally, we study the impact of the switch architecture on our provisioning by restricting our approaches to full-wavelength MDDRs.
© 2010 Optical Society of AmericaFull Article | PDF Article