Input queued switches: cell switching vs. packet switching

Input Queued (IQ) switches have been very well studied in the recent past. The main problem in the IQ switches concerns scheduling. The main focus of the research has been the fixed length packet-known as cells-case. The scheduling decision becomes relatively easier for cells compared to the variable length packet case as scheduling needs to be done at a regular interval of fixed cell time. In real traffic dividing the variable packets into cells at the input side of the switch and then reassembling these cells into packets on the output side achieve it. The disadvantages of this cell-based approach are the following: (a) bandwidth is lost as division of a packet may generate incomplete cells, and (b) additional overhead of segmentation and reassembling cells into packets. This motivates the packet scheduling: scheduling is done in units of arriving packet sizes and in nonpreemptive fashion. In M.A. Marsan et al. (2001) the problem of packet scheduling was first considered. They show that under any admissible Bernoulli i.i.d. arrival traffic a simple modification of maximum weight matching (MWM) algorithm is stable, similar to cell-based MWM. In this paper, we study the stability properties of packet based scheduling algorithm for general admissible arrival traffic pattern. We first show that the result of Marsan et al. extends to general regenerative traffic model instead of just admissible traffic, that is, packet based MWM is stable. Next we show that there exists an admissible traffic pattern under which any work-conserving (that is maximal type) scheduling algorithm will be unstable. This suggests that the packet based MWM will be unstable too. To overcome this difficulty we propose a new class of "waiting" algorithms. We show that "waiting"-MWM algorithm is stable for any admissible traffic using fluid limit technique.