TLEN 5830 Nuno Monteiro

1
TLEN 5830Nuno Monteiro

• An analysis of Quality of Serivce Schemes For
  IEEE 802.11 Wireless LANs by Lindgren, Almquist,
  Schelen

2
Introduction

• Background Information on wireless vs wired
  networks
• Wireless has become dirt cheap!!
• Comparison with previous work and how they
  extended the research Past work just mentioned
  the schemes but did not test how they work in
  real world
• Discuss the need for service differentiation
• Some applications require more priority
• The four new schemes that will be examined 1)PCF
  2)EDCF 3)DFS and 4) Blackburst.
• THESIS Of these 4 which gives the best
  performance (defined in the metrics section) for
  High Priority nodes
• Methodology and Scope
• 1) Overview of 802.11 DCF. 2) Discuss how the 4
  schemes work. 3) Discuss the simulation and the
  metrics used to compare them. 4) Compare and
  analyze the Results for each scheme.

3
Overview of DCF

• Basic Access mechanism of IEEE 802.11 which uses
  CSMA/CA
• How it works
• Sense the media for a DIFS period
• If media is idle send the frame else
• Once the media becomes idle for a DIFS period,
  decrement your backoff time (B) until its the
  nodes turn to Tx.
• B is chosen randomly from 0 - Contention Win
• If before the nodes B reaches zero, someone else
  begins to Tx then pause the decrement of B and
  only begin to decrement B once the media has
  become idle for a DIFS
• If a collision occurs the node must double its CW
  and choose a random backoff .

4
Overview of PCF

• Centralized polling based scheme which coexists
  with DCF.
• How it works
• Time divided into a superframe where there are
  two parts to a superframe
• 1)Contention Free Period (CFP) where PCF is used
  by some nodes. Length of CFP may vary
• Started by a special frame (beacon) from the PC
• During CFP the Point Cordinator (PC) (which has a
  cache of which stations want to access using PCF)
  tells which station when it can Tx.
• Each Tx during the CFP must not be separated
  longer than a PIFS period which is lt DIFS
• 2)Contention Period where DCF is used by the
  other nodes that are not using PCF. Length of CP
  must be at least the max length of a frame.

5
Overview of EDCF

• A Scheme in which priority is given depending on
  the 8 different traffic classes that a node is
  in.
• It is backwards compatible with DCF
• How it works
• It gives shorter CWmin for different stations
• Stations w/ the same CWmin get priority by
  getting its own AIFS (extra time slot)
• Each station therefore waits a DIFS period plus
  AIFS. The less AIFS the more priority a station
  will get.
• It allows for packet bursting. Once a node gets a
  hold of the media it can Tx for a predefined max
  time (TxOpLimit).
• To prevent others from Tx during your bursting
  period you must ensure that the amount of time
  inbetween your packets is no longer than a SIFS

6
Overview of DFS

• A Centralized scheme in which a single node
  always has info about all flows
• How it works
• A weight is assigned to each different
  Transmission (flow).
• This weight affects the B period
• Before Tx each node must wait its B period.
• Formula
• If collision occurs the CWmin is set to 3.
• Such a short CWmin so that each flow can be Tx as
  soon as possible.
• Each flow has a different weight attributed to
  it, the assumption is that it would be rare for a
  collision to occur

7
Overview of Blackburst

• Goal is to min delay for real time traffic
• Distinction b/t High vs Low priority nodes
• All high priority (HP) stations try to access the
  medium w/ constant intervals and are able to jam
  the media. Low priority(LP) nodes cant jam
• How it works
• When HP nodes want to send a frame it must sense
  the media for a PIFS, if media is idle it Tx its
  frame else the media is busy
• If Media is busy the HP begins to calculate
  increase the amount of blackslots it will Tx.
• Amt of blackslots(very short time interval) is
  dependent on how long the HP has been waiting to
  access the media. The longer the wait the more
  blackslots it Tx. Once the media has been idle
  for a PIFS it begins to Tx its blackslots (ie
  Jamming the media)
• If the node is the last one jamming the media it
  Tx within a PIFS period
• LP nodes use ordinary DCF to access the media

8
Simulations and Scenarios for test

• Network simulator ns-2 w/ DCF functionality and
  where they added the other 3 schemes
  functionality
• The tested scenarios specifics
• Simulations modeled a 2Mbps WLAN.
• Each station communicated with a Base station.
  That was in turn wired with 10Mbps
• Each station could hear all the other stations.
• They tested each service differentiation scheme
  with changes in the amt of high and low priority
  nodes.
• They started with few HP nodes and slowly raised
  the number of HP nodes to see the effect on
  metrics that they tested for.
• They tested low bit rate vs high bit rate

9
Metrics for evaluation

• There are five metrics that they tested
• Average throughput shows how well the schemes
  differentiate b/t various priority levels
• Normalized throughput allows comparison b/t
  different levels of load.
• Medium utilization shows the efficient use of
  the scarce wireless spectrum.
• Collision Rate average number of collisions per
  second
• Access delay see the delay attributed to the
  data.

10
PCF long or short superframe

• Since the size of the superframe for PCF is not
  defined the group first tested PCF to determine
  what is the best size for a superframe.
• Done through testing avg throughput and medium
  utilization the result was that a short
  superframe (20TU) was best for HP nodes although
  LP nodes suffered.
• RESULTS Access Delay
• The best scheme in minimizing delay is blackburst
  while the worst seems to be DFS. However all
  schemes have relatively acceptable and similar
  delays.

11
Results - Throughput

• Blackburst demonstrates that it is the best
  overall option. In low bit rate as more HP nodes
  are added performance drops in the other schemes
  before it does for Blackburst
• only when there is high bit rate along with many
  nodes does Blackburst drop below the performance
  of other schemes
• In terms of HP nodes DFS drops in performance
  before other schemes
• Normal for DFS because it attempts to provide
  fair differentiation for LP and HPs

12
Results Med. Utiliz. collisions

• Medium Utilization
• Where there is low load blackburst is the best
  for Med. Utilization.
• When there is high load blackburst is the best
  until the number of nodes gets high (over 15)
  then it begins to deteriorate. Due to the fact it
  avoids collisions so well (due to black slots)
  overhead increase greatly deteriorating its util.
• Collisions per second
• Blackburst has the least amt of collisions when
  there are few LP stations no matter how many HP
  stations there are. EDCF is the worst.
• When there are many LP nodes and few HP nodes
  then DFS is the best but as HP stations increase
  Blackburst becomes the best in minimizing
  collisions. EDCF is the worst.

13
Good points of the paper

• Methodology
• Scope
• They remained true to their research by choosing
  the option (i.e. short length of superframe for )
  where service differentiation was maximized in
  the respective scheme.
• In some of these options LP nodes were neglected

14
Issues that were not clear

• The paper failed to discuss several items that it
  mentioned
• The delay portion in the PCF scheme
• The random variable in the DFS
• Arbitrary assignment of 20TU for a short frame
  and over 50TU for a long frame
• Didnt say why 1TU 1024microseconds
• The section on the simulator
• Their ability to implement blackburst, EDCF and
  DFS into the ns-2 simulator.
• Their decision of replicating real world
  scenarios while it wasnt clear why the criteria
  that they choose exactly matched the real world