Control architecture in optical burst-switched WDM networks

On the optimality of packet-oriented scheduling in photonic switches with delay lines

“…As a result, wavelength conversion and optical buffering can (and should) be integrated consistently in a single contention resolution strategy. This integration has been referred to earlier as wavelength allocation in optical buffers [3,26,28], later as the channel and delay selection (CDS) problem [5], and is also the topic of the current paper.…”

“…Due to this, on a given wavelength, it is in general not possible to schedule a packet/burst just after the transmission of a previous one. Therefore, on each outgoing wavelength, gaps occur, which are time periods during which no transmission takes place on the outgoing wavelength, even though packets/bursts are awaiting transmission in the optical buffer [25][26][27][28]. These periods are also referred to as voids, and can be considered as a strict capacity loss, with an effect that can be best compared to an increase in traffic congestion [1,10].…”

“…Furthermore, it has poor scalability, since the hardware complexity grows along with the maximal number of packets/bursts that can be processed simultaneously. For these reasons, a horizonbased algorithm is preferred, maintaining only real value per channel, namely the scheduling horizon [3,4,6,15,22,25,27,28], and is also assumed here.…”

“…The FDL buffering architecture and related CDS control problem is identified already in [9]; however, the control algorithms of [25,28] are probably best-known. In [25], Turner introduces horizon scheduling, a scheduling algorithm which bases its decision upon the horizon values of the available channels.…”

“…As an example, consider the two CDS algorithms known from literature: MINL (minimum length) (introduced in [28] and named MINL in [3]) and MING (introduced in [3]). Both are burst-size-independent, and do not involve preventive dropping.…”