Enhancing SWAN QoS Model
By Adopting Destination-Based Regulation (ESWAN)
Agenda
|
|
|
Ad-Hoc QoS Independent Models |
|
|
|
SWAN Building Blocks |
|
|
|
SWAN Dynamic Regulation Issues |
|
|
|
ESWAN Basic Terms |
|
|
|
Observations On SWAN Model |
|
|
|
ESWAN Destination-Based Regulation |
|
|
|
Evaluation of ESWAN |
|
|
|
Conclusion |
|
|
Ad-Hoc QoS Independent
Models
|
|
|
|
FQMM (IntServ + DiffServ) |
|
dQoS (Fair Dynamic Bandwidth
Percentages) |
|
INSIGNIA (Stat-full Model) |
|
SWAN |
|
Stateless |
|
AIMD |
|
Pragmatic Conservative Expectation of
Bandwidth |
|
No Resource Allocation |
|
Distributed |
|
Soft QoS Guarantees |
|
E2E RT-Delays ??? Effective BW ??? |
SWAN Building Blocks
|
|
|
|
|
Admission Controller |
|
|
|
Classifier |
|
|
|
Shaper |
|
|
|
Rate Controller |
|
|
SWAN Admission Controller
(AC)
|
|
|
Conservative look at BW availability. |
|
Use MIN BW available between SRC/DST. |
|
Slack of BW used by BE. |
|
Use of probe req. reply. |
|
Admission provides soft guarantees
only. |
|
Intermediate nodes do not allocate
resources. |
SWAN Packet Classifier
|
|
|
Simply forwards unmarked packets
to the Shaper |
|
Forward marked packets to the MAC layer |
SWAN Rate Controller (RC)
|
|
|
Follows conservative AIMD approach. |
|
Limit throughput to the knee of AIMD
chart. |
|
Relies on information from MAC layer
(packet delay). |
SWAN Dynamic Regulation
Issues
|
|
|
Congestion Due to Mobility |
SWAN Source-Based
Regulation
|
|
|
Congested node must mark all passing RT
packets with CE. |
|
Destination nodes inform source nodes
with the congestion via QoS reporting mechanism. |
|
Source nodes slows down transmission
rates. |
|
If new rate is unsatisfactory, source
nodes re-establish the QoS flow. |
|
Fast recovery, aggressive and
disruptive to many QoS flows. |
SWAN Network-Based
Regulation
|
|
|
Congested node selects only subset of
passing RT-flows (victim flows). |
|
Same consequences as in Source-Based |
|
If congestion is not resolved in time T,
select new (victim flows). |
|
Victim flows are selected randomly. |
|
Slower recovery, less aggressive to
RT-flows. |
|
|
ESWAN Basic Terms
|
|
|
Define MAPD (Max Acceptable Packet
Delay) |
|
Define Effective BW = BW without
expired packets. |
|
Define Effective BW Ratio |
|
Define Limited and High QoS |
|
Define Effective Delay Ratio |
Observations On SWAN
Model
|
|
|
About 10% of RT-packets arrive expired. |
|
No parameters to measure flow QoS. |
|
No mechanism designed to respond to
degraded QoS. |
|
Expired BW consumes network resources. |
ESWAN Destination-Based
Regulation
|
|
|
|
|
Preemptive Behavior |
|
|
|
Performed by DST node on flows
experiencing Limited QoS (no congestion). |
|
Limited QoS condition is triggered when
(b < bL)T. |
|
DST node issues regulate message to SRC
node. |
|
SRC node performs re-initiate procedure
to locate new (better) route. |
|
Preemptive behavior is a preventative
measure. |
|
Preemptive behavior maintains QoS
levels. |
|
|
ESWAN Destination-Based
Regulation
|
|
|
|
|
Recovery Behavior |
|
|
|
Congested node marks all RT-packets
with CE until node realizes sufficient decrease in arriving BW. |
|
DST nodes having (d > d1)
will issue regulate message. |
|
Other DST nodes wait for time T. If
packets keep arriving with CE, nodes having (d > d2) will
regulate. |
|
After another time T, if packets keep
arriving with CE, nodes having (d > d3) will regulate. |
|
Values (di) are network
constants, and (di > di+1). |
Evaluation of ESWAN 1
|
|
|
Preemptive behavior limited amount of
expired BW in ESWAN compared to SWAN. |
|
Recovery behavior caused smoother
recovery from congestion in ESWAN compared to SWAN. |
Evaluation of ESWAN 2
|
|
|
Preemptive behavior caused limited
effect on BE traffic. |
|
Recovery behavior caused a much less
congested node in ESWAN compared to SWAN. The recovery is smoother in ESWAN. |
Evaluation of ESWAN 3
|
|
|
The delay chart of RT-packets in ESWAN
shows less congested packets and fewer expired. |
|
Cumulative RT-packet delay % in ESWAN
shows about 1.2% only of delivered packets are unusable. |
Evaluation of ESWAN 4
|
|
|
Higher EBR in ESWAN due to preemptive
behavior. |
|
Lower EDR in ESWAN due to recovery
behavior. |
|
Better NW utilization in ESWAN. |
Conclusion
|
|
|
Preemptive behavior decreases chances
of congestion. |
|
|
|
ESWAN facilitates a way to upgrade
provided services. |
|
|
|
Gradual recovery less disrupting to
RT-flows. |
|
|
|
Shaping Rate T is empirical value (2
sec). |
|
|
|
MAPD is a flow specific value (known to
DST). |
|
|
|
EBR (b) & EDR (d) are important QoS
measures. |
|
|
|
Larger ad-hoc NW need the MAPD to limit
expired BW. |
|
|
|
Future investigation on the effect of
imbalanced load bet. RT and BE traffic in SWAN enabled ad-hoc NW. |
Questions & Remarks