A Distributed Efficient Flow Control Scheme for Multirate Multicast Networks

1
A Distributed Efficient Flow Control Scheme
forMulti-rate Multicast Networks

• Reporter Naixue Xiong

2
Outline

• Context Motivation
• Flow control system model
• PID controller
• Stability analysis and control gain selection
• Algorithm implementation
• Simulation results
• Conclusion

3
Outline

• Context Motivation
• Flow control system model
• PID controller
• Stability analysis and control gain selection
• Algorithm implementation
• Simulation results
• Conclusion

4
Context Motivation

• Definition Multicast is a communication
  referring to distributing data from a source to
  multiple destinations.
• MotivationA major factor of traffic on the
  Internet with rapid growth of multimedia and
  other data distribution applications.
• The most widely used multicast
  transport protocols, which are layered on top of
  the IP multicast layer, can cause congestions or
  even congestion collapses if they do not provide
  adequate congestion control.
• Considerable research efforts focused
  on congestion control schemes for multicast
  services.

5
Context Motivation

• Classification Rate-based (this )
  Window-based (AQM, RED, )
• Choice Rate-basedWindow-based scheme has extra
  complexity in maintaining and synchronizing the
  congestion window across all receivers.It
  usually generates data bursts periodically.
• ClassificationSingle-Rate Multicast
  (SR-M),Multi-Rate Multicast (MR-M).

6
Outline

• Context Motivation
• Flow control system model
• PID controller
• Stability analysis and control gain selection
• Algorithm implementation
• Simulation results
• Conclusion

7
The System Model
Figure 1. a multi-rate multicast model, where
every router can have downstream routers or
end-users
8
The System Model
xi, buffer occupancy of RTi , target buffer
occupancy of RTi ti, delay from RTi' to RTi ti
, round-trip delay between RTi and RTiIi,
expected incoming rate to RTiOi, maximum
outgoing rate of RTi to downstream routers
Figure 2. Router RTi, its incoming data rate from
its upstream router RTi' and its outgoing data
rate
9
Outline

• Context Motivation
• Flow control system model
• PID controller
• Stability analysis and control gain selection
• Algorithm implementation
• Simulation results
• Conclusion

10
the error signal between thebuffer occupancy and
the target buffer occupancy at time slot n
PID rate controller
sum of RTi's history sending rates during the
last round-trip time ti
the differential signal of the buffer occupancy
between the current time slot and last slot.
(n) , target buffer occupancy at time slot n
Ii(0), the initial incoming rate to RTia, bt
(t1, 2, ti ) and c are the proportional,
integral and derivative parameters,
respectively.
11
A Digital Filter feedback control process
derivative component
integral component
Proportional component,Initial incoming rate
12
Outline

• Context Motivation
• Flow control system model
• PID controller
• Stability analysis and control gain selection
• Algorithm implementation
• Simulation results
• Conclusion

13
Steps

• Apply z-transformation
• z-domain representation
• Characteristic Polynomial (CP)
• all poles of CP in the open unit disk of the
  complex plane
• Results
• Theorem 1

14
Steps

• Apply z-transformation
• z-domain representation
• Characteristic Polynomial (CP)
• all poles of CP in the open unit disk of the
  complex plane
• Results
• Theorem 1

15
Steps

• Apply z-transformation
• z-domain representation
• Characteristic Polynomial (CP)
• all poles of CP in the open unit disk of the
  complex plane
• Results
• Theorem 1

16
Steps
Necessary to choose the appropriate parameters,
such that all poles of CP lie inside a unit
disk.

• Apply z-transformation
• z-domain representation
• Characteristic Polynomial (CP)
• all poles of CP in the open unit disk of the
  complex plane
• Results
• Theorem 1

17
Steps

• Apply z-transformation
• z-domain representation
• Characteristic Polynomial (CP)
• all poles of CP in the open unit disk of the
  complex plane
• Results
• Theorem 1

18
Outline

• Context Motivation
• Flow control system model
• PID controller
• Stability analysis and control gain selection
• Algorithm implementation
• Simulation results
• Conclusion

19
Intra-session Fair

• Intra-session Fairness Algorithmallocate
  different rates to different users, which
  ensures good intra-session fairness.
• It is fully distributedno need of global
  knowledge
• It is fair to each receiverrate received by
  receivers depends on the capacity
• Buffer occupancy be stabilized quickly,
  consideration change within round-trip time.

20
Inter-session Fairness

• Inter-session Fairness Algorithmdifferent
  multicast sessions share the link bandwidth in a
  fair fashion.
• It is fair to all multicast sessionsavoids
  starvation of sessions that require less
  bandwidth.
• It has high bandwidth utilizationfully utilized
  by all multicast sessions
• is simple and highly efficientrate computed by
  the simple formula (18)

21
Outline

• Context Motivation
• Flow control system model
• PID controller
• Stability analysis and control gain selection
• Algorithm implementation
• Simulation results
• Conclusion

22
Performance Evaluation
23
As time goes on, when the BCPs arrive at the
sources, the multicast sources adjust the sending
rate gradually to stabilize the traffic at all
routers.
the sending rates of the sources S1 and S2 are
stabilized at the rates of 5 Mbps and 4 Mbps.
Fig. 6 shows the sending rates of multicast
sources S1 and S2, and CBR source SCBR, which
corresponding to multicast sessions 1, 2,
3, respectively, in the simulation figures.
24
There are some fluctuations at the beginning
because of network delay and response of the PID
controllers. The PID controllers quickly adjust
the expected rates for downstream routers and
users.
Fig. 7. The fair rates of Session 1, Session 2,
and Session 3 in link L1
Fig. 8. The fair rates of Session 1 and Session 2
in link L11.
25
Fig. 9. The fair rates of Session 1, Session 2,
and Session 3 in link L12. Fig. 10. The fair
rates of Session 1, and Session 3 in link L121.
26
The ratios of RT1, RT11, RT12 and RT121 gradually
stabilize at 92, 86, 95, and 91 at 121 ms,
120 ms, 98 ms and 95 ms, respectively.
Fig. 11. The ratios of buffer occupancy to RT1
and RT12. Fig. 12. The ratios of buffer
occupancy to RT11 and RT121.
27
Fig. 13. The link utilizations of link L1 and
link L11. Fig. 14. The link
utilizations of link L12 and link L121.
For link L11, there are only multicast flows, so
the link utilization stabilizes at 92. For links
L1, L12, L121, because the CBR flow passes
through them, there are some fluctuations.
Nevertheless, the link utilizations are
eventually stabilized around 82, 88, and 83,
respectively.
28
From the Figs. 15-16, there are some fluctuations
at the beginning, while our control scheme
quickly stabilizes the buffer occupancy. We can
find that there are still slight fluctuations in
stable range for RT1.
How the parameters inuence system stability
29
Figs. 17-18 give the change of the ratios of
buffer occupancy for RT1, RT12, RT11 and RT121
with e 1/(t i10). It is clear that there are
slight fluctuations for RT1 and RT12 during the
simulation time. When e 1/(ti10), its
stabilization period is more erratic than that
with e 1/(ti6).
30
We find all the four curves have slight
fluctuations during the simulation time. It
performs worse in terms of control than the above
three cases, leading to a waste of resources.
31
Outline

• Context Motivation
• Flow control system model
• PID controller
• Stability analysis and control gain selection
• Algorithm implementation
• Simulation results
• Conclusion

32
Conclusion

• presented an efficient flow control schemeusing
  an explicit rate feedback mechanism.
• proposed a PID controller to stabilize the buffer
  occupancy at routers, and the traffics in the
  network can thus be stabilized.
• control theory to determine the PID parameters to
  ensure the stability of the control loop.
• superior performance of our scheme in
• Simulation demonstrated performance in terms of
  system stability, link utilization, and
  intra-session and inter-session fairness.

33
Context Motivation

• Two classifications of AQM
• Queue-based which controls the queue at the
  congested link
• e.g., RED (random early detection) and most of
  its variants
• Rate-based which controls the flow rate at the
  congested link
• e.g., AVQ (adaptive virtual queue algorithm)