Title: Bandwidth Reallocation for Bandwidth Asymmetry Wireless Networks Based on Distributed Multiservice A
1Bandwidth Reallocation for BandwidthAsymmetry
Wireless Networks Based onDistributed
Multiservice Admission Control
- Robert Schafrik
- Lakshman Krishnamurthy
2Agenda
- Introduction
- Related work on Admission control and bandwidth
allocation - Distributed Multiservice Admission Control
- System model DMS-AC in Two-cell system
- DMS-AC in Multi-cell system
- Performance evaluation
- Competing Systems
- Static Allocations
- Conclusion
- Comments
3Introduction
- Next generation multiservice wireless networks
are expected to present distinctive traffic
asymmetry between uplink and downlink. - Some resources may be wasted if bandwidth is
allocated symmetrically - To match the asymmetric traffic load, it is
necessary to allocate different bandwidth to
uplink and downlink. - Different call classes have different up/down
ratios - QoS may be different for Handoff and new call,
and for each call class
4Introduction (continued)
- If the traffic and mobility patterns are
predictable, then fixed bandwidth allocation
works. - Bursty and variable bandwidth requirements call
for new treatments of network resource management - Traffic generated is time dependent
- It is necessary to develop a dynamic bandwidth
allocation scheme that can adapt to the changing
traffic conditions
5Problem Statement
- Upload and Download communications are not always
symmetric - Need to determine under what conditions bandwidth
needs to be reallocated - Need to determine the best way to reallocation
when multiple call classes and multiple cells
while preserving QoS
6Time Slots
- Some timeslots are for uplink, some are for
downlink. This prevents collisions - Variable time-slots for different cells always
outperforms fixed time slots - Reallocation of time slots affects all calls in
the system, try to limit how frequently this is
done
7QoS Metric
- Call Admission Control (CAC)
- Critical CAC Parameters
- Pn New call blocking probability
- Ph Handoff call blocking probability
- MINBlock used to optimize
8Distributed Multiservice Admission Control
(DMS-AC)
- Provides a base to compare new techniques against
- Tries to find proper threshold
- Limits new calls of certain classes
- If blocking probability exceeds a bound, it
reallocates - If QoS thresholds for some classes cannot be
found, it reallocates
9Related Work
- CAC schemes
- CDMA (fixed, symmetric)
- CDMA/TDD (fixed, asymmetric)
- SA same-slot allocation (all cells have same
allocation) - DA different slot allocation (cells can have
different allocations, but adjacent cells may
have slot interference) - Limited Fractional Guard Channel scheme
- DCA Distributed Admission Control
- Jeons CAC for MSWN 7
- DMS-AC scheme
10Limited Fractional Guard Channel (LFGC)
- Minimize a linear objective function
- Weighted sum of handoff and new call blocking
probabilities - C channels
- C-T reserved for new and handoff
- When T channels are used, only handoff calls are
accepted - Extended to deal with multiple call classes20
11Distributed Admission Control (DCA)
- Based on communication between cells to predict
handoffs - Only deals with one call class
- Knapsack problem 18 to deal with multiple call
classes
12Distributed Multiservice Admission ControlSystem
Model
- Total bandwidth allocated of a cell is fixed.
- Bandwidth allocated on uplink and downlink is
different and also adjustable 3 8 - M classes of calls in the system
- The calls of particular class have the same
bandwidth requirements, mobility characteristics
and mean resource holding time
13Distributed Multiservice Admission ControlSystem
Model (contined)
- Design goal of the proposed admission control
scheme is - fi lt ?i
- Fi lt ?i
- ?i (eta) - Highest tolerable dropping
probability of class i hand-off calls. - fi (phi) hand-off call dropping probability of
class i calls - Fi (phi) New call blocking probability of class
i calls. - ?i (rho) Highest tolerable new call blocking
probability
14Distributed Multiservice Admission ControlSystem
Model (contined)
- DMS-AC operates in distributed manner
- System states exchanged periodically between
adjacent cells - Base station of cell makes an admission decision
based on the state information of the cell itself
and its neighboring cells. - DMS-AC uses the admission threshold of each call
class based on the system states to limit the
admission of new calls. - Dynamic threshold scheme is used.
- Threshold of specific call class is recomputed
and reset periodically. - Control period interval between two threshold
computing process (15 60 minutes).
15Distributed Multiservice Admission control in a
Two-Cell System
- Fig. 1. Two-cell system.
- Cr is the observing cell and Cl is the
neighbouring cell - Total bandwidth in Cr (Cl) is denoted by Bru
Brd (Blu Bld) - In DMS-AC we need to define the overload states
of a specific call class in the multiservice
system. - In multiservice networks, the set of overload
states of different call classes may be different.
16Distributed Multiservice Admission control in a
Two-Cell System (contined)
- Example
- Cell has 10 downlink and 5 uplink channels
- Class 1 calls require 1 uplink and 1 downlink
channel - Class 2 calls require 1 uplink and 3 downlink
channels - (n1, n2) denote the system states, where n1 and
n2 denote the class 1 calls and class 2 calls in
the system - (0,3) and (2,2) are overload states of class 2
calls. No class 2 calls are not admissible while
class 1 calls are admissible.
Fig. 2. An example. (a) Overload states of class
1 calls. (b) Overload states of class 2 calls
17Distributed Multiservice Admission control in a
Two-Cell System (contined)
- During a control period, the admission of class i
new call in the observing cell Cr should satisfy
the following two conditions - The admission of a new class i call in Cr cannot
cause the call dropping probability of class j
call in Cr denoted by frj to exceed ?j - The admission of a new class I call in Cr cannot
cause the call dropping probability of call class
j in the neighboring cell Cl, denoted by flj to
exceed ?j - The key of DMS-AC is to determine the thresholds
of individual call class in each cell (i.e. we
need to compute frj and flj)
18Distributed Multiservice Admission control in a
Two-Cell System (continued)
- The key of DMS-AC is to determine the thresholds
of individual call class in each cell (i.e. we
need to compute frj and flj) - The probability that xi class i calls out of ri
calls stay in Cr has a binomial distribution
given by - Similarly, the probability that yi class i calls
handoff to Cr from Cl during the control period
is
19Distributed Multiservice Admission control in a
Two-Cell System (continued)
- Using formulas 1 2, we need can find Pr(ni)
- Pr(ni) denote the probability that there are ni
class i calls in Cr during T units of time - At any time system stays in feasible state,
should satisfy
20Distributed Multiservice Admission control in a
Two-Cell System (continued)
21Distributed Multiservice Admission control in a
Two-Cell System (continued)
- Blocking probability of class j calls in Cr can
be expressed as - Blocking probability of class j calls in Cl is
expressed as
22Derivation of Admission threshold
- Thi1 and Thi2 denote the thresholds of class i
calls that satisfies the first and second
admission conditions - The final admission threshold of class i calls in
Cr, which satisfies all admission conditions , is
given by
23Extension to multicell system
- C0 be the current observing cell
- C1 to C6 be the neighboring cell
24Extension to multicell system
- During a control period, the admission of a class
i (I ? 0, M-1) call in C0 should satisfy - The admission of a new class i call in C0 cannot
cause the call dropping probability of call class
j in C0, f0j to exceed ?j - The admission of a new class i call in C0 cannot
the call dropping probability of call class j in
the neighboring cells to exceed ?j
25Valid States
q
n
- Number of calls for class q
- in the system
M
Number of feasible states Constrained by Bu and
Bd
Call classes
Not all of these states are good for the system,
but they are possible. Matrix will not be
symmetric.
S(i,j) is the subset of states such that adding a
call of type i will cause overload for class j
26Threshold-Based Admission Control Scheme
If you are the current call class (note not
always zero!)
Test for conditions 1,2, and 5
You are NOT the current cell
Conditions
27Case 1 Cell i will Become Full for Some Call
Class
- si is in the set S(i,j) adding a call type i
will cause at least one other class j to become
full - Need to reallocate up/down channels
28Admission Case 1 Ratio of Uplink and Downlink
Needs to Change
Need to choose an allocation between
29Admission Case 2 Cell r Will Not be able to
Accept Handoff from Cell l
- Cell r either doesnt have enough room or
accepting a handoff will cause a class to
overload - See if Cell r can reallocate to accommodate
30DMS-AC Pseudocode
31Comparisons
- Analysis using a 2-cell system
- 15 minute control period
- 100 channels
- 2 call classes
- Real time ( 1 up, 1 down )
- Non Real Time ( 1 up, 3 down )
32Jeons scheme
- Similar goal create a scheme for reallocating
in asymmetric environments - Accounts for traffic load in both directions
- Uses Markov analysis
- Also only considers QoS for New and Handoff calls
33Comparison with Jeon (1)
- New call QoS is similar, and not shown
- Jeon does not consider NRT QoS
34Comparison with Jeon (2)
- Call types vary independently
35Comparison with Jeon (3)
- Similar performance with small loads
- Jeons begins to lag with NRT calls
- Jeons breaks down when volume is high
36Comparison with Static Soln (1)
- No reallocation is performed for AC without BA
- RT call arrival rates in both Cr and Cl increase
from 0.07 to 0.12 simultaneously - Up/down ratio is 30 up/ 70 down
37Comparison with Static Soln (2)
- average NRT call arrival rates in both Cr and Cl
change from 0.006 to 0.011 simultaneously - Up/down ratio is initially 50 up/ 50 down
38Comparison with Static Soln (3)
- Traffic increases for Cr, decreases for Cl
39Conclusions
- Changing the up/down ratio for several asymmetric
call classes helps maximize the resources of a
Cell, and still guarantees QoS for new and
handoff calls - When to reallocate
- Allocations for a call class nears max
- Allocations of neighbor for that class nears max
- How to reallocate
- Find min B that fills QoS requirement
40Comments
- Experimental setup was simplistic
- Perhaps more than 2 call types could be
considered in a simulation - Perhaps compare the performance of more than 2
cells - A call cannot itself be dynamic (aka use
1up/1down for a while then switch to 1up/2down) - Does not consider revenue, but that might be
achievable by adjusting admittance thresholds - Performs slightly better than Jeon in some
conditions
41Backup Slides
42Other Notes
- Assume C0 covers a conference, and becomes
overloaded - C1 - C6 will be unable to accept any calls of any
class (due to the handoff constraint)
C0
43Overview
- Uplink and Downlink bandwidth is asymmetric
- Determine when to change ratio of uplink to
downlink - Determine how to compute best ratio to satisfy
QoS - Satisfy QoS for call classes