Is Random Access Fundamentally Inefficient

1
Is Random Access Fundamentally Inefficient?

• Rajmohan Rajaraman
• Northeastern University

2
What is Random Access?

• Each node accesses the channel independently with
  a certain probability
• Fundamentally a symmetry-breaking technique
• Different protocols vary according to how this
  probability is chosen
• Fixed (e.g., Aloha)
• Fixed with carrier sense (e.g., basic CSMA
  protocols)
• Varies dynamically including collision avoidance
  and backoff (e.g., 802.11 DCF)
• What is not random access?
• Coordinated access to the channel (e.g., TDMA)
• Using information from higher layers (e.g., flow
  rates, interference)

3
Single-Hop Scenario

• All nodes can hear one another
• Uniform rate r for each of n links -- nr lt 1
• Random access with fixed probability can achieve
  throughput within 1/e of optimal (Aloha)
• For nr close to 1, TDMA clearly more efficient
• Arbitrary rates ri with ? ri lt 1
• Random access with uniform probability clearly
  inefficient
• Random access with rate-dependent probabilities
  can achieve throughput within 1/e of optimal
  (Chafekar et al 2008)
• Again, as the aggregate rates come close to 1,
  scheduled schemes outperform

4
Dynamic Link Rates

• Random access determines access probabilities
  based on channel contention
• Backoff schemes
• For low aggregate rates, backoff methods
  efficient
• In theory, exponential backoff schemes converge
  much slowly and are unstable for even small
  aggregate rates bounded away from 1 (Leighton et
  al 88)
• Polynomial backoff schemes -- contention window
  polynomial in no. of collisions -- more effective
  (Goldberg-Mackenzie 97)
• Poor average delay
• In the absence of information about dynamics,
  random access with backoff probably competitive

5
Multihop Networks

• Random access (in its basic form) susceptible to
  interference problems
• Information available from channel different for
  different users sharing the same channel
• Fundamentally inefficient (cannot address
  hidden/exposed terminals)
• Random access also oblivious to end-to-end flow
  control
• Access probabilities based on channel contention
  may be inefficient and inconsistent with flow
  control
• The above problem is not with the random part
  of random access -- it is due to the lack of
  information

6
Informed Random Access

• A joint scheduling/rate control/routing
  optimization scheme computes local link rates
  Tassiulas-Ephremides 92, Cruz-Santhaman 03, Jain
  et al 03
• Takes interference into account
• End-to-end flow control and packet routing
• A very complex problem, but orthogonal to whether
  random access is used at the MAC level
• Each node made aware of rates at which adjacent
  links would be used
• Random access based on these link rates Yi et al
  07
• In some sense reduces multihop case to single-hop
  case
• Random access efficient if aggregate channel
  rates low Joo-Shroff 07
• Possibly competitive when link rates vary highly
  dynamically

7
In Closing

• The answer depends on
• The information available to the MAC layer and
  whether random access uses this information
• Whether joint rate control/routing optimization
  being done at higher layers
• Saturated vs unsaturated conditions
• Kind of traffic (bursty vs constant-rate)
• Scenarios where random access could be
  competitive
• Unsaturated (load in all parts of the multi-hop
  network should be well below saturation)
• Highly dynamic environment where joint
  optimization is impractical
• Random access ineffective when
• Saturated conditions
• Access does not take into account
  interference/routing/flow info
• Static non-uniform traffic patterns much better
  handled by scheduling
• Need delay guarantees

8
What was Missing

• Power control and energy efficiency
• Security and resiliency to DoS attacks
• Fairness
• Specific QoS issues