Multicast Scheduling of Wavelength-Tunable, Multiqueue Optical Data Center Switches
AffiliationUniv Arizona, Coll Opt Sci
Optical packet switching
MetadataShow full item record
PublisherOPTICAL SOC AMER
CitationKamran Keykhosravi, Houman Rastegarfar, and Erik Agrell, "Multicast Scheduling of Wavelength-Tunable, Multiqueue Optical Data Center Switches," J. Opt. Commun. Netw. 10, 353-364 (2018)
Rights© 2018 Optical Society of America
Collection InformationThis item from the UA Faculty Publications collection is made available by the University of Arizona with support from the University of Arizona Libraries. If you have questions, please contact us at firstname.lastname@example.org.
AbstractThe all-optical switching of multicast flows using star couplers and tunable transceivers is a promising solution for emerging cloud data center applications. However, the limited tuning range of optical components on one hand and the buffer management challenges for multicast traffic delivery on the other pose a significant impact on the performance of optical multicast scheduling algorithms. Using only one queue per input port results in head-of-line (HOL) blocking and limits the throughput, especially for bursty traffic patterns. As the number of possible multicast destinations grows exponentially with the switch size, allocating one queue per destination is not a feasible solution. To resolve HOL blocking, in this paper we consider only a handful of queues per switch input port and devise scalable scheduling algorithms that take into account transceiver tunability constraints. According to our Monte Carlo analysis of a switch with 64 ports and operating under bursty traffic, it is possible to improve the maximum achievable throughput by 44% when the number of queues per port is increased from one to eight. We show that the performance gains due to an increase in the queue count depend on the availability of the spectral resources. With the scarcity of wavelengths, an increase in the number of queues leads to diminishing returns.
Note12 month embargo; published online: 15 March 2018
VersionFinal accepted manuscript
SponsorsSwedish Research Council [2014-6230]; NSF Center for Integrated Access Networks (CIAN) [EEC-0812072]