Title: Implementation Of A Testbed For Evaluating Energy Efficiency In Passive Optical Network
1Implementation Of A Testbed For Evaluating
Energy Efficiency In Passive Optical Network
- Supervisor Submitted by,
- Dr. Luca Valcarenghi Neelakandan Manihatty Bojan
2Outline
- Introduction
- Scope of the thesis
- State of the art
- Testbed architecture
- Testbed implementation
- Conclusion
- Future scope
- Bibliography
3Introduction
4ONU architecture
5Stratix 4GT FPGA board
6FPGA architecture
7Scope of the thesis
- To implement an Energy Efficient ONU (ECONU)
prototype for EPON in Altera FPGA based on the
finite state machine (FSM).
8State of the art
9Testbed architecture
10Frame and Packet format
11VLSI design flow
Design Specification
Behavioral Description
RTL Desctiption (HDL)
FunctionalVerification and Testing
Logic Synthesis
Gate-Level Netlist
LogicalVerification and Testing
Floor PlanningAutomatic Place Route
Physical Layout
Layout Verification
Implementation
12Birds eye view of the design
13Negative exponential inter-arrival time
14Birds eye view of sleep controller module
15FSM implementation
16Resource utilization
Device resources summary
Resource utilization of design
17SignalTap
18Testbed setup
Simulation Synthesis
Programming the FPGA
.sof (SRAM based object file)
Commands given through System console and Signal
tap
Executes commands and provides the results back
19Results
There are two main test modes used in the design
Internal Loopback
Local Loopback
20Sleep time
21Low power idle time
22Wake time
23Analytical evaluation
- Sleep time Ts 2.88µs
- Low power idle time Tlp 39.68µs
- Wake time Tw 4.48µs
- Total sleep time Actual sleep time
Overhead associated with transitions - Low power idle time
Sleep time Wakeup time -
39.68µs 2.88µs
4.48µs -
47.04µs - So the overall percentage of the actual sleep
time in a sleep request - (Low power idle time / Total sleep
time)100 - (39.68µs/47.04µs)100
- 84.35
- Similarly the overall percentage of the sleep
time and the wake time is 6.1 and 9.5
respectively.
24Energy efficiency characteristics
Packet size 1500 Bytes
Packet size 1000 Bytes
25Energy Efficiency characteristics
Packet size 500 Bytes
Packet size 100 Bytes
26Analytical evaluation of energy savings
- Pa Power consumed by the ONU during Active
state - Plpi Power consumed by the ONU during Low
Power Idle state - Ta Time spent in Active state
- Ts Time spent in Sleep state
- Tlpi Time spent in Low Power Idle state
- Tw Time spent in wake state
- ONUcycletime SONUtimespentinindividualstates
Ta Ts Tlpi Tw - Let the energy consumed by the ONU with sleep be
denoted by Es and the one without Sleep technique
be Ews - EwsPa(Ta T s Tlpi Tw)
- The energy consumed by the ONU when it is has
the Sleep technique implemented in it, is given
by the following equation. - EwsPa(TaTs Tw) PlpiTlpi
- Energy Savings ((Ea-Eb)/Ea)100
- Energy Savings ((Pa-Plpi)Tlpi/(Pa(TaTs
Tlpi Tw)))100 - Consider Pa 10 Plpi
27Generator consistency
Throughput Vs Packet size
28Exponential inter-arrival time results
29Conclusion
- This thesis proposed and implemented a sleep mode
technique in the data link layer for Ethernet
PONs in Altera Stratix 4GT FPGA. The sleep
technique is based on the FSM implementation
written in Verilog HDL. - The FSM switches the ONU on and off based on
the sleep request en-coded explicitly through out
green header implementation based on the sleep
time, low power idle time and wake time are all
based on IEEE 803.az standard. - The analytical evaluation for the proposed
design provides Energy efficiency of 75 , when
the power during active state is ten times the
power at low power state. The results obtained
after the implementation coincides with the
analytical results. Hence substantial energy
savings is achieved with our implementation. - The implementation was performed on the testbed
setup in two different modes local loopback and
internal loopback. Similar results were obtained
for both the configurations. For the
implementation with standard values as defined in
IEEE 803.az - Additionally, a negative exponential
inter-arrival module was developed. This is
integrated in the generator to emulate the inter
packet arrival times. So that packet arrivals can
be modeled with various statistical distributions.
30Future scope
- For the ONU to have low power consumption, it has
to remain in the sleep mode as long as possible.
The longer the better. But for guaranteed quality
of service (QOS), the buffers should be large
enough to store all the packets which arrive even
when the ONU is sleeping. - They buffers should be intelligently managed
both in terms of buffer size and the number of
buffers. This in itself is an interesting task
which requires buffer optimization mainly due to
the limited resources available in the FPGA. - In the real time scenario, implementation of
the buffer(s) would be inevitable because - Numbers of users are ever increasing.
- Traffic is increasing.
- Bandwidth requirements are rising.
- QOS has to be guaranteed.
- Taking all the above into considerations, a
dynamic sleep protocol is required. This should
not only manage the buffers, but also should
synchronize with both the OLT and ONU.
31Grazie