Title: QoS
1QoS
Bandwidth Allocation
for
EPON
Final Presentation
Nadav Aharony
Eyal Nave
Project Advisors Nir Naaman and Shay Auster
2Project Goal
Design and test bandwidth allocation
algorithms for the EPON upstream
- Design and analyze several bandwidth allocation
algorithms - Create EPON simulation in OPNET
3Project Stages
Scheduling Algorithms
OPNET Simulation
Project Definition
Proposals and Fine-tuning
Prototype
GOAL Design and test bandwidth allocation
algorithms for the EPON upstream
Basic Scenario (Phase 1)
Analysis
Comparison
Algorithm Simulation
Conclusion
4EPON Background
EPON Ethernet over Passive Optical Networks
IEEE 802.3ah Ethernet in the First Mile (EFM)
Task Force
Current initiative defines the fiber access rate
at 1Gbps Future expansions to 10Gbps
Central Office (CO) OLT Optical Line
Terminal Customer Premise (CP) ONU Optical
Network Unit
Preliminary Approximation for a 1Gbps Link 32
64 ONUs
5Bandwidth Allocation Principles
- Upstream shared link is managed by the OLT
- ONU will transmit only in its allocated grant
slot - A minimal grant will be allocated for each ONU in
each cycle to allow it to send bandwidth
requests. - There is no random-access in EPON
6OPNET Implementation
- Creation of a completely new environment in
OPNET - A working OLT device
- A working ONU device
- Implementation of relevant 802.3ah MAC control
commands. - Implementation of the TDMA timing regime
- In addition, the following were implemented for
the Prototype - Implementation of a static allocation scheme
- Basis for expansion to the more complex
algorithms. - Implementation of CBR based Ethernet sources.
- A test scenario to help in debugging and provide
proof that the system works as expected.
7Prototype Scenario
Grant Allocation Messages
Downstream Fiber
Upstream Fiber
TDMA ONU Traffic Status Reports
8Modeling TDMA
Cycle N
Cycle N1
Cycle N-1
Cycle N-2
next cycle start
OLT TIME"
do schedule
allocation algorithm at work
send grants
REPORT for cycle N
"ONU TIME"
3
2
1
9 New MAC Messages
REPORT
GATE
- GATE
- Start Time
- Length
- Generate Report?
REPORT Status of 8 Priority Queues
10ONU Data Path
Subscriber Traffic
Prioritized Traffic Msgs
To PON
From PON
1 Gbps
ONU Traffic
MAC Control Msgs
ONU Data
11ONU Data Path
Prioritized Traffic Msgs
Subscriber Traffic
To PON
From PON
MAC Control Msgs
1 Gbps
ONU Traffic
ONU Data
12ONU Scheduler
Schedule Grant Interrupts
Future control msgs
Start Transmition Actions
13ONU Queue
Transmition Initiation Actions
Single Packet Transmition
Idle While In GATE
Idle While Not In Gate
Packet Insertion
Generate Status Report
14OLT Data Path
To PON
GATE Msgs
MAC Control Msgs
From PON
1 Gbps
Upstream Data
15OLT Data Path
GATE Msgs
To PON
MAC Control Msgs
From PON
1 Gbps
Upstream Data
16OLT Scheduler
Collect and handle reports
Future control msgs
Compose GATE messages based on Grant Table and
send to Network
Insert Algorithm Here!
17EPON Prototype Test Results
Queue behavior compared to grant sizes
Small Gates
Medium Gates
Large Gates
End-to-End Delay vs. the number of active ONUs
18EPON Prototype Test Results
End-to-End Delay vs. the number of active ONUs
19Bandwidth Allocation and Analysis
- Proposal of allocation algorithms
- Analytical analysis of the algorithms
- Algorithm implementation in OPNET
- Construction of appropriate simulation scenarios
and the conduction of the simulations - Construction of more advances traffic sources,
network traffic and network user profiles - Analysis of simulation results
20Problem Definition
- Two types of network traffic
- Committed Rate (CR)
- Best Effort (BE)
- Fairness definitions
- Dont ask, dont get
- History window accounting
- All ONUs with similar history info will be
treated equally
21Allocation Algorithms
- 1. Static Allocation
- GrantCycle_Size/(Number_of_ONUs)
Cycle N
Cycle N1
Cycle N-1
- 2. Semi-Static Allocation
- Pre-defined allocations according to boolean
requests from ONUs. - All ONUs are given a minimal grant to allow for
B/W requests. - Rest of bandwidth is equally split between ONUS
with requests gt 0
3 requests
2 requests
1 request
Example
22Allocation Algorithms
23Allocation Algorithms
- The grants are a function of
- ONU Requests
- Service Level Agreement (SLA)
- Fairness in the allocation process among ONUs
2416 ONU Scenario Definitions
- Network sources generate Ethernet packets
- Packet length is an exponential with mean of 3000
bits - Packets with length of over 1500 bytes are
discarded (Ethernet MTU size). - Packet inter-arrival time is exponentially
distributed
Average source bit-rate is determined through the
setting of different inter-arrival mean values.
High Load 100 Mbps Medium Load 50
Mbps Low Load 5 Mbps
x 8 ONUs
x 4 ONUs
x 3 ONUs
x 1 ONUs
Idle 0 Mbps
Total 16 ONUs
25Simulation Timeline
- Total Simulated time 0.69 sec
- Half of ONUs from each load type defined as
stable same source behavior throughout the
simulation - Rest of ONUs defined to have a traffic burst
during the simulation. - Cycle size set to 500 µsec ? 1380 cycles
simulated
PON Average Bitrate Mbps
Four main time segments
1015
2
545
510
3
4
1
0
0.3
0.5
0.69
tsec
0.05
26Sample Results
27Sum of Granted Bits
High Load ONU
Medium Load ONU
Low Load ONU
Queuing Delay
High Load ONU
Medium Load ONU
Low Load ONU
28Sum of Granted Bits
80,000kbit
50,000kbit
100,000kbit
High Load ONU
Medium Load ONU
Low Load ONU
Queuing Delay
High Load ONU
Medium Load ONU
Low Load ONU
0.001sec
0.0004sec
0.25sec
29Sum of Granted Bits
High Load ONU
(100 Mbps)
- Static allocation 30 Mbps
- ? Constant under-granting
- ?A 100 Mbps ONUs queue would explode
- Segment 1
Medium Load ONU
Low Load ONU
High Load ONU
Queuing Delay
Low Load ONU
30Sum of Granted Bits
High Load ONU
(100 Mbps)
- Static allocation 30 Mbps
- ? Constant under-granting
- ?A 100 Mbps ONUs queue would explode
- Segment 1
- Segment 2
Medium Load ONU
Low Load ONU
High Load ONU
Queuing Delay
Low Load ONU
High Load ONU
Medium Load ONU
31Sum of Granted Bits
High Load ONU
(100 Mbps)
- Static allocation 30 Mbps
- ? Constant under-granting
- ?A 100 Mbps ONUs queue would explode
- Segment 1
- Segment 2
- Segment 3
Medium Load ONU
Low Load ONU
High Load ONU
Queuing Delay
Low Load ONU
High Load ONU
Medium Load ONU
32Sum of Granted Bits
High Load ONU
(100 Mbps)
- Static allocation 30 Mbps
- ? Constant under-granting
- ?A 100 Mbps ONUs queue would explode
- Segment 1
- Segment 2
- Segment 3
- Segment 4
Medium Load ONU
Low Load ONU
High Load ONU
Queuing Delay
Low Load ONU
33Sum of Granted Bits
80,000kbit
50,000kbit
100,000kbit
High Load ONU
Medium Load ONU
Low Load ONU
Queuing Delay
High Load ONU
Medium Load ONU
Low Load ONU
0.001sec
0.0004sec
0.25sec
34Sum of Granted Bits
Low Load ONU
(5 Mbps)
80,000kbit
50,000kbit
100,000kbit
- Static allocation 30 Mbps ? Constant
over-granting - Puzzle! Semi-Static should not allocate LESS
than Static for a stable source! - ExplanationPacket arrival rate is lower than
REPORT rate, resulting in cycles with no grant
for the ONU - Also accounts for delay difference between Static
and Semi-Static - Note Dynamic Algorithms delays
High Load ONU
Medium Load ONU
Low Load ONU
Queuing Delay
High Load ONU
Medium Load ONU
0.001sec
0.0004sec
0.25sec
35Sum of Granted Bits
80,000kbit
50,000kbit
100,000kbit
High Load ONU
Medium Load ONU
Low Load ONU
Queuing Delay
High Load ONU
Medium Load ONU
Low Load ONU
0.001sec
0.0004sec
0.25sec
36High Load Burst ONU
- At 0.3 sec, source drops from 100Mbps to 5Mbps,
and shuts down at 0.5 sec. - Note how each algorithm handles the burst
37Sum of Granted Bits
Dynamic Algorithm
Semi-Static Algorithm
Static Algorithm
38Sum of Granted Bits
Dynamic Algorithm
- Individual handling of each ONU type
- Continued increase of allocated bits due to high
volume of REPORT during times of low congestion
- Low-load 5Mbps ONUs are negligible
Dynamic Algorithm
Semi-Static Algorithm
Static Algorithm
39Sum of Granted Bits
Semi-Static Algorithm
- Medium-load and high-load ONUs converge
- Similar slopes indicate similar allocations
- Idle ONU not allocated any bandwidth
Dynamic Algorithm
Semi-Static Algorithm
Static Algorithm
40Sum of Granted Bits
Static Algorithm
- Same allocation for all ONUs as expected
Dynamic Algorithm
Semi-Static Algorithm
Static Algorithm
41Conclusions
- For heterogeneous-source networks ?
- dynamic algorithm achieves best partition of
network bandwidth - Of the three algorithms, the dynamic algorithm
also handles high-load bursts with best results
42Conclusions (Cont.)
- For ONUs with low load sources
- in a situation of a high loaded network ?
- The dynamic algorithm shows the worst delay
performance out of the three algorithms. - - The other algorithms allocate more than the low
load ONU needs, which is utilized for sending
newly-arrived packets that were not reported - Other ONUs in the network may experience
increased delays because of the wasted bandwidth.
43Conclusions (Cont.)
- For the same reason
- Static allocation provides better delay results
for the low-load ONUs than the semi-static
algorithm since it keeps allocating bandwidth
even if it does not receive a request for it. - The downsides of the Static algorithm are clear
- Wastes bandwidth that may be needed by other
ONUs - Prevents over-subscription to the network.
- Sets a top limit on the allocated bandwidth per
ONU
44Conclusions (Cont.)
- The Semi-Static algorithm
- For ONUs that request less bandwidth than the
semi-static algorithm eventually allocates them ?
Same downsides as the Static allocation - Handles situations with idle ONUs well.
45The End