Comparison of

a packet switching versus cell switching

for input queuing Switches
Reviewed by :

Nahum Setu
BDU1101864
 Acknowledgments

• I would like to express my grate gratitude to my advisor,

Pushparaghavan Annamalai (Ph.D)for being my advisor and
his constant encouragement, support, and invaluable
suggestions made this work successful.

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 2
Switches
 Motivation

• Scheduling is relatively easier when all packets have fixed length for the
same cell size.
• However, in practice packets are not segmented into fixed length,
therefore such kind of scheduling in the case of cell-based switching it
comes with some disadvantages:
I. Loss of bandwidth due to the existence of incomplete cells, and
II. Additional overhead of segmented packets and re-assembly of cells
• So those problems strongly motivate me to study packet-based
scheduling, which is the transfer of packets without segmenting them.

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 3
Switches
 Overview

• Introduction

• Literature Review

• Objective

• Statement of problem

• Methodology

• System modeling

• Results and Discussion

• Conclusions

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 4
Switches
 Introduction

• Input queue  (IQ) - is a processes of scheduling a packet that are waiting

to be sent from input ports to output ports.
• Input Queued switch architecture has low memory bandwidth
requirements compared to other known architectures,
• Cell switching- is a process of scheduling a variable packet length which is
segmented into fixed length.
• Packet switching- is a process of using a packet directly without breaking
them into cells.

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 5
Switches
 An input-queued switch

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 6
Switches
 Literature Review

• Packet Scheduling in Input-Queued Cell-Based Switches –This paper show a maximum

throughput achieved by input-queued packet switches is identical to that achieved with

input- and output-queued cell switches. [M. A. Marsan, et al.]

• Achieving 100% throughput in an input-queued switch- This paper show longest queue

first can lead to the permanent starvation of short queues. oldest cell first overcomes

this limitation by favoring cells with large waiting times.[N. McKeown, et al.]

• The throughput of data switches with and without speedup- This paper show use fluid

model techniques to establish a maximum weight algorithm for connecting inputs and

outputs delivers a throughput of 100%.[J. G. Dai and B. Prabhakar]

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 7
Switches
 Objective

• General objective
– To show Packet-based algorithm is stable for arbitrary admissible
arrival processes,

• Specific objective
– To propose a new class of “waiting” algorithms and

– To show that “waiting” PB-MWM algorithms is stable for any

admissible traffic using the fluid model technique.

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 8
Switches
 Statement of problem

• Packet-based maximum weight Matching (PB-MWM) is rate stable for any

IID (independent and identically distributed) admissible traffic with IID
packet lengths having finite mean and variance.
• But, if packet lengths are not IID, a packet-based scheduling algorithm, in
particular PB-MWM, is not always stable.
• Hence, in this presentation we will see different type of packet-based
scheduling algorithm in order to obtain stability for an arrival processes.

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 9
Switches
 Methodology

• Fluid model

– fluid model of a switch operate under some matching algorithm m

which satisfy traffic arrival process for a large number N.
– The stability of the fluid model under MWM algorithms use the
stability of discrete time switch.

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 10
Switches
 Conti..

• Let be the number of cells in Virtual Output Queue sat time ,

including any arrival at time
• Let denote the number of cells that have arrived at input destined to
output j up to time.
• Let be the number of departures From up to time .
• The first equation simply states that the number of cells in at any time equals
the total number of arrivals minus the total number of departures.
• The second equation states that the number of departures till time n equals the
number departure still previous time slot (time slot n-1) plus the departures at
time n.

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 11
Switches
 System modeling

• Packet Based-Waiting Maximum Weight Matching (PB-wMWM)- Algorithm At each time slot,

the MWM algorithm selects the matching with the maximum weight among all matchings in M.

If there are multiple such matchings, one of them is selected arbitrarily.

– When the input and output ports are free packet wait for an indefinite number of time slots.

– When a packet gets served, do not schedule the freed ports till all ports become free and

schedule according to a full MWM schedule.

– The waiting synchronizes the weight of the schedule to the weight of the MWM schedule.

– Hence, if waiting is done frequently enough, we can verify that the weight of the schedule is

not more than a bounded constant away from MWM

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 12
Switches
 Results and Discussion

• However, we note that during the waiting period some ports lose
bandwidth.
• Hence, if waiting is done too aggressively then the algorithm can not
utilize the full bandwidth.
• These observations lead to the following waiting algorithm which we
denote as PB-wMWM.
• PB-wMWM algorithm is stable under any admissible traffic with a known
bound on maximum packet length with max speed.

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 13
Switches
 Conclusions

• Modification of A cell-based MWM for packet scheduling yields 100%

throughput for any admissible for IID traffic with independent packet lengths.

• But, if packet lengths are not IID, a packet-based scheduling algorithm, in

particular PB-MWM, is not always stable.

• To overcome this problem we proposed a new class of waiting algorithms.

• Under the waiting algorithm the switch becomes stable for any admissible

traffic.

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 14
Switches
 Reference

1. M. A. Marsan, A. Bianco, P. Giaccone, E. Leonardi, and F. Neri,

“Packet scheduling in input-queued cell-based switches,” in Proc.
IEEE INFOCOM, 2001, pp. 1085–1094.
2. N. McKeown, V. Anantharam, and J. Walrand, “Achieving 100%
throughput in an input-queued switch,” in Proc. IEEE INFOCOM,
1996, pp. 296–302.

3. J. G. Dai and B. Prabhakar, “The throughput of data switches with

and without speedup,” in Proc. IEEE INFOCOM, 2000, pp. 556–564.

Comparison of a ps vs cs for input queuing

Monday, January 10, 2022 15
Switches