Infrastructure for CrossLayer Designs Interaction

1
Infrastructure for Cross-Layer Designs Interaction
Zhijiang Chang, Georgi Gaydadjiev, Stamatis
Vassiliadis Computer Engineering
Laboratory Electrical Engineering, Mathematics
and Computer Science Delft University of
Technology
2
Outline

• Introduction
• Problems of Standalone CL Designs
• Previous CL Architecture Proposals
• Proposed Solution
• Generic collaboration architecture
• Dedicated priority control mechanism
• Cross-layer designs optimization guidelines
• Minimal overhead signaling and explicit encoding
• Simulation Results
• Delay Optimization
• Throughput Optimization
• Conclusion and future work

3
Cross-layer Designs for MANET

• Wireless ad hoc network protocols are based on
  wired network properties
• Cross-layer designs aim at different optimization
  targets, e.g. for
• Application QoS control MAC application
  layers
• TCP fake congestion avoidance MAC TCP layers
• Global traffic balancing MAC IP TCP layers.

4
Cross-layer for TCP Fake Congestion Avoidance
example

• Package loss causes
• network congestions, or
• link errors.
• Wrong Assumption for MANET the package loss is
  caused only by network congestion (in wired NW,
  the Bit Error Rate lt 0.001)
• In MANET the wireless link errors dominate
  (interference, obstacles, out-of-sight (low
  SNR)).
• Possible solution MAC layer information used to
  determine the exact reason.

5
Standalone CL Designs deficiencies

• Single cross-layer aims at limited subset of the
  QoS metric
• Multiple cross-layer designs are required to
  achieve full QoS optimization
• compatibility problem coexistence causes
  dependency loop, etc.
• interaction problem priority control among CL
  designs, etc.

6
CL Design shortcomings due to

• Faulty assumptions caused by incorrect or
  incomplete information
• Shared cross-layer data strict R/W policy,
  priority control
• Not intended dependency loops (bad trigger
  conditions)
• Additional overhead internal as well as external
  overhead (e.g. CrossTalk).

7
Previously proposed architectures
.

• Mobileman
• Central network entity (Network Status) for CL
  information
• complex R/W control
• Huge storage requirements (e.g. harddisk)
• The layered architecture is removed completely.

• ÉCLAIR
• Function call based interface in Tuning Layers
• Cross-layer design interactions not supported
• Highly OS dependent.

8
Our Solution

• Overall QoS control
• Maintain layered protocol stack structure
• Support interaction priority control
• Allow cross-layer designs refinement
• Provide efficient signaling and universal
  understandable encoding.

9
Collaboration Architecture
Our Architecture
Desired model
The structure should take care of the global QoS
requirement and consequently the overall
performance.
The control system that supports priority
handling of cross-layer designs is located in the
MAC layer .
10
Interaction Priority Control

• Two types of interactions
• between CL designs
• between the middleware CL design.
• Four priority levels (from high to low)
• The control middleware
• CL designs with global (network) knowledge
• CL designs for point-to-point (MAC, PHY, IP)
• CL designs for end-to-end (TCP, APP).

Registration
Interaction command
11
Guidelines for CL designs

• Three phases should be supported
• Activation phase CL design registration
• Decision making phase Read information and take
  local decisions (no writing actions)
• Action phase Based on the priority arrangement,
  write/update the parameters.

12
Signaling and Encoding
Universal understandable encoding - XML
Efficient low cost signaling
Basic expression
lttype_of_infogt ltnamegtvaluelt/namegt ltownergtname_of_o
wnerlt/ownergt ltlvlgtlevel_of_ownerltlvlgt lt/type_of_in
fogt
The parameter example
ltvargt ltTCP_SWgt128lt/TCP_SWgt ltownergtTCP_CNG_AVlt/owne
rgt ltlvlgtlocal_highltlvlgt lt/vargt

• The information is piggybacked on normal
  communication
• Minimized overhead (variable information size
  used)

The control command example
ltcntlgt ltnamegtSUPPRESS_TCP_CNG_AVlt/namegt ltownergtCro
ssTalklt/ownergt ltlvlgtgloballtlvlgt lt/cntlgt
13
Experimental Results

• Simulation Infrastructure
• Based on Network Simulator 2 (NS-2) ver. 2.28
• OLSR package from University of Murcia UM-OLSR
• Simulation Scenario
• Mobility model random waypoint
• Node speed 0-20m/sec
• Number of nodes 50
• Area 500m X 500m
• Traffic sources
• 10 CBR on UDP
• 25 FTP on TCP
• Test duration 600 sec.
• Routing Protocols
• Reactive Ad hoc On-demand Distance Vector (AODV)
• Proactive Optimized Link State Routing (OLSR)

14
Dedicated CL Optimizations

• Three CL optimizations
• Integrated MAC/PHY layer (Opt. 1)
• Priority Local Lower-layer
• Local wireless media related information SNR
  etc. for other CL designs
• TCP congestion optimization. (Opt. 2)
• Priority Local Higher-layer
• Fade congestion avoidance
• Global load balancing optimization (Opt. 3)
• Priority Global/Network
• Network situation evaluation with one-hop
  neighbors link information
• Combinations to observe interaction between Opt2
  Opt3.
• COM1 Opt.1 Opt.2
• COM2 Opt.1 Opt.3
• Our Proposal Opt.1 Opt.2 Opt.3

15
Delay Optimization
OLSR
AODV

• The delay optimization mainly happens when pause
  time lt120s
• COM1 has negative impact of delay when network
  is unstable, while COM2 can avoid that.
• Our proposal can suppress COM1 using information
  from COM2.

16
Throughput Optimization
OLSR
AODV

• The throughput optimization mainly happens
  when pause time gt120s
• Our proposal benefits from both COM1 COM2
• Our proposal can suppress COM2 with the overhead
  information provided by COM2 itself (OLSR pause
  time gt 30).

17
Results Summary

• All interactions verified (between CL designs and
  between CL designs control middleware)
• The architecture does not degrade the performance
  of the original protocols or individual CL
  designs
• The proposed architecture has the particular
  weaknesses inherited from individual CL designs.

18
Conclusion Future Work

• Negative impact of individual cross-layer design
  can be foreseen and avoid
• Multiple cross-layer designs cooperate according
  to the priority rules
• Our proposal is universal, OS independent and
  highly modular
• The control middleware only lies in the MAC
  layer, easier to implemented in SW or HW.
• Continue research on the global QoS control
  algorithm
• Improve the credibility of source information
• Investigate the internal overhead
• Implement this architecture for MISAT project.

19
(No Transcript)