Weighted Fair Queueing

1
Weighted Fair Queueing
GPS PGPS SCFQ Implementation
2
Round-robin
a
a
e
b
b
e
c
d
c
d
(a)round-robin
(b) weight round-robin

• problems
• packets lengths are not the same ? not fair

3
Generalized Processor Sharing (GPS) (A. Parekh,
1992)

• assumptions
• traffic is infinitely divisible.
• serve multiple sessions simultaneously based on
  the ratio of pre-defined weights.
• properties
• fairness
• minimum throughput guarantee
• analyzable with leaky bucket admission control

4

• notations
• an example of GPS

5

• minimum throughput guarantee
• Summing over all session j, service rater

6
PGPS (A. Parakh, 1992)

• PGPS (packet-by-packet scheme) an approximation
  to GPS for operating in real world
• ideally, PGPS wants to serve packets in the
  increasing departure order of GPS
• an example of different between GPS and PGPS

7

• properties of PGPS
• departure time of the two scheme

8

• properties (continued)

9

• Hard to keep track of departure time (GPS)
• implementation of PGPS (virtual time)
• use hypothetical ideal fluid flow model as
  reference
• use virtual time, v(t), to represent the progress
  of work in the reference system.
• example

10
(No Transcript)
11

• PGPS algorithm
• virtual time finishing times can be determined at
  the packet arrival time
• packets served in order of virtual time finishing
  time
• problems
• not feasible
• way of assigning weight is not defined

12

• picking up

13
Self Clocked Fair Queueing J. Golestani(1994)

• Motivation
  
  eliminate the need for the
  hypothetical fluid-flow reference system.
• Approach
  different
  notion of virtual time, which depends on the
  progress of work in the actual packet-based
  system.

14

• SCFQ algorithm
• each arriving packet is tagged with a
  service tag.
• service tags are iteratively computed as
• virtual time is defined equal to the service tag
  of the packet receiving service.
• transmit packets in the increasing order of
  service tags.

15

• Properties of SCFQ
• Implementable
• fairness property
• The end-to-end session delay bounds are
  comparable to that of PGPS with leaky bucket
  admission policy.
• Problems
• Way of determining value is still not
  defined.

16
Efficient FQ Architectures(info 96)

• Motivation
• develop efficient scheduling algorithms for
  high-speed ATM network
• Contents
• sorting scheme
• hierarchical approach for wide range of services

17

• SCFQ
• To simplify implementation, calculate the service
  tag until the cell reaches the head of its
  connection queue.

18

• Observation
• for session k,

19

• Architecture
• if the ith bin is receiving service, put the new
  head-of line cell to the
  th bin
• logic to transmit a cell and locate the next
  non-empty sorting bin is required.
• problem
• a large number of sorting bins are required to
  handle a wide range of bandwidth parameters

20

• Hierarchical Fair Queueing (two-level scheduler)
• first stage handle connections with similar rates
• second stage serves a small number of groups
• Group weights