CATV Networks Bandwidth Allocation for the Upstream Channel

1
CATV Networks Bandwidth Allocation for the
Upstream Channel
Final presentation (winter 2001)
A project by Chen Yaniv 031763071 Lavi
Abramovich 032214363
Instructor Nir Menkerman
2
Project Definitions .
Learning and Understanding the DOCSIS/MCNS
standard for data over cable networks.

Developing an efficient scheduling algorithm for
the Cable Modem Termination System(CMTS).

Demonstrate the behavior of the algorithm with a
simulator implemented in Matlab.

3
Why Cable Network ?(Physical Layer)
Transparent IP Traffic, Through The
Data-Over-Cable System
- Economic

• Broadband communication
  service.

- Widespread
- Integrate all communication services (Video,
Data)
4
Bandwidth Allocation MAC Layer
The upstream channel is modeled as a stream of
mini-slots.
The CMTS allocates the upstream slots by
broadcasting a map message to the CMs.
Mini slots
Previous map
current map
Yet unmapped time
5
Requests

• The map includes allocation of the following
  types of upstream messages for the CMs (SIDs)
• Unicast (data or request) supports piggyback
• Multicast (requests)
• Data/Request immediate access
• contention resolution based on exponential back
  off .

6
Map Properties
7
CMTS Transportation Policy
QoS demands for Min/Max bandwidth, Timing

8
Best Effort
We focused on developing the algorithm for
supplying the regular users, the best bit rate
possible.
9
Solution The Algorithm
Parameters T simulation total time W window
time Delta - map size (mini slots) MinBW MaxBW Win
dowPriorities PendingVector RequestVector
10
Algorithm Improvements

• Support concatenation.

• Use map remains.

• Best decision between modems with
• the same priority.

11
Performance Evaluation

• Compare with FIFO algorithm
• Find 100 capacity a-posteriori.
• Delay count.

12
Implementation IssueHow to Model the Cms ?

• Does the modem's requests affected by the CMTS
  allocation policy?

2) What kind of random variables would best
represent the modems behavior ?
3) Can a modem support fragmentation ?
13
Our SolutionUse Buffers
FIFO Algorithm
GetRequestsBuff()
Fairness Algorithm
14
Simulation Results (1)

• 200 modems.
• All the modems are the same.
• No concatenation support.
• Simulation time 2000 sec
• Window size 4 sec
• Map size 60ms (2400 minislots)

15
Simulation Results (2)

• 40 modems.
• Linear increasing cms requests
• No concatenation support.
• Simulation time 2000 sec
• Window size 4 sec
• Map size 60ms (2400 minislots)

16
Simulation Results (3)

• 200 modems.
• Linear increasing cms requests
• 100 supports concatenation.
• Simulation time 2000 sec
• Window size 4 sec
• Map size 60ms (2400 minislots)

17
Simulation Results (4)

• 40 modems.
• Linear increasing cms requests
• No concatenation support.
• Simulation time 2000 sec
• Window size 4 sec
• Map size 60ms (2400 minislots)

18
Conclusions

• Fairness was achieved!
• Channel capacity usage is best when map size is
  much bigger then average request length.
• Table statistics and all other algorithm
  improvements are most efficient when requests are
  small in compare to the map size.
• Delay is minimized.

19
Future Suggested Improvements

• Optimization of concatenation max size and
  W/delta ratio.
• Add protocol overhead.
• Support dynamic map size.
• No need in table when all requests can be
  allocated without pending.