Designing a TCP/IP Core for Power Consumption Analysis

1
Designing a TCP/IP Core for Power Consumption
Analysis

• Kenichi Tanamachi, Koji Inoue, and Vasily G.
  Moshnyaga
• Fukuoka University

2
Background

• TCP/IP protocol designed by software
• Major protocol for network data transactions
• Network bandwidth increasingly grows
• Grow 1Gbps in near future
• Bottleneck in high-speed network
• Low-speed TCP/IP packet processing
• of termination nodes
• Require low-power for mobile network-connected
  computing devices
• PDA, Cellular phone,

3
TCP/IP protocol stack
Transmit
Receive
Application layer(HTTP,SMTP,POP3)


Data link layer (Ethernet,PPP)
Data
Data
Data
TCPheader
Data
TCPheader
Data
TCPheader
IPheader
IPheader
Data
TCPheader
IPheader
Ethernet header
Ethernet header
Internet
4
Purpose
Implement TCP/IP packet processing by custom
hardware and analyze its power consumption

• In order to analyze power consumption
• Step1 Decide TCP/IP Core organization
• Step2 Design TCP/IP Core
• Step3 Analyze the power consumption

5
Design Flow

• Functional partition
• -decide functional partition of TCP and IP
• HDL
• -RTL design by using Verilog-HDL
• Logic synthesis
• -use synopsys Design Compiler with 0.35µm CMOS
  library
• -output report file for area and power
  consumption
• Layout
• -use an auto PR tool
• -extract wire load capacitance
• Analysis

HDL
Simulation
Library
Logic Synthesis
Report file
Layout
6

• TCP Core

7
Functional Partition
Application layer

• port_ctr (BLOCK1)
• Transfer data to or from application layer
• data_ctr (BLOCK2)
• Divide and reconstruct data by using sequence
  numbers
• Error check by timer
• window_ctr (BLOCK3)
• Congestion control by inform transmit-side
  receive-sides empty buffer size
• checksum (BLOCK4)
• Check error of data and header, and if have some
  error, create the request of retransmission

TCP
port_ctr
data_ctr
window_ctr
checksum
IP layer
Data link layer
8
Result(Area)
Transmit-side
Receive-side
?Note Since TCP functions are different between
transmit-side and receive-side, the measurements
have been done by generating each circuit
9
Result(Power consumption)
Transmit-side
Receive-side
10
Result(Wire capacitance)
Transmit-side
Receive-side
11
Design Results
Area Area Power Power Wire capacitance Wire capacitance
transmit receive transmit receive transmit receive
BLOCK1 4 6 6 4 4 3
BLOCK2 27 23 23 30 22 26
BLOCK3 7 10 10 7 9 6
BLOCK4 62 61 61 59 65 65
Checksum in BLOCK4 requires a large number of
transistors
12

• IP Core

13
Functional Partition
Application layer

• sevice_info (block1)
• Inform header information( IPversion, header
  length, packet length, type of service)
• frag_ctr (block2)
• Determine whether to fragment or not data. If
  yes, identify each data
• add_ctr (block3)
• Determine source and determination address
• Checksum (block4)
• Check error in IP header and compute time to live
  for packet

TCP layer
IP
service_info
frag_ctr
add_ctr
checksum
Data link layer
14
Result

• Power consumption

• Area

15
Design Results
Power
Area
16
14
block1
29
26
block2
43
47
block3
12
13
block4
Checksum in block3 requires a large number of
transistors
16
Discussion

• Which block gives a great impact on power?
• TCP Core
• Checksum(BLOCK4) More than 60 of all
• IP Core
• Checksum(BLOCK3) More than 47 of all

Focus on these two blocks to lower the power
consumption of TCP/IP Core
17
Comparison of TCP with IP
TCP
IP
receive
transmit
3.332
2.1196
3.774
Area(µ?)
19.669
10.7182
18.8503
Power(mW)
The TCP Core includes more complex functions
than the IP Core