The Role of

1
The Role of Awareness inInternet Protocol
Performance

• Carey Williamson
• iCORE Professor
• Dept of Computer Science
• University of Calgary

2
Introduction

• It is an exciting time to be an Internet
  researcher (or even a user!)
• The last 10 years of Internet development have
  brought us
• World Wide Web (WWW)
• Media streaming applications
• E-commerce, eBay, e-government
• Peer-to-peer (P2P) applications
• Wi-Fi wireless LANs
• Mobile/pervasive/ubiquitous computing

3
Theme of this Talk

• Existing layered Internet protocol stack does not
  lend itself well to providing optimal performance
  for diversity of service demands and environments
• Who should bend users or protocols?
• Explore the role of awareness in Internet
  protocol performance
• Identify tradeoffs, evaluate performance

4
Internet Protocol Stack

• Application supporting network applications and
  end-user services
• FTP, SMTP, HTTP, DNS, NTP
• Transport end to end data transfer
• TCP, UDP, RTP, SCTP, XTP
• Network routing of datagrams from source to
  destination
• IPv4, IPv6, BGP, RIP, routing protocols
• Data Link frames, channel access, flow/error
  control
• PPP, Ethernet, IEEE 802.11b
• Physical raw transmission of bits

001101011...
5
Viewpoint

• Layered design is good
  layered implementation is bad
• Good
• unifying framework for describing protocols
• modularity, black-boxes, plug and play
  functionality, well-defined interfaces (SE)
• Bad
• increases overhead (interface boundaries)
• compromises performance (ignorance)

6
Main Example TCP

• The Transmission Control Protocol (TCP) is the
  protocol that sends your data reliably
• Used for email, Web, ftp, telnet,
• Makes sure that data is received correctly right
  data, right order, exactly once
• Detects and recovers from any problems that occur
  at the IP network layer
• Mechanisms for reliable data transfer
• sequence numbers, ACKs, flow control, timers,
  retransmissions, congestion control...

7
In Praise of TCP

• TCP is the 4 wheel drive of transport layer
  protocols
• general purpose, robust, go anywhere
• The TCP protocol has undergone only minor changes
  in 30 years of existence
• original version circa 1974
• congestion control mechanism 1988
• TCP has witnessed dramatic changes in network
  technology and in computing technology over that
  same time period!

8
Criticisms of TCP

• TCP is a high overhead protocol
• extra network packets for handshaking
• extra RTTs for handshaking
• Performance problems aplenty
• high delay-bandwidth product networks
• multiple packet losses in same window
• phasing effects
• fairness problems
• wireless networks
• World Wide Web

9
The Problem Restated
TCP or not TCP? That is the question!
- William Shakespeare (1608)
10
TCP Performance Problems

• Examples
• TCP over ATM Networks
• TCP over Wireless Networks
• TCP over Wireless Ad Hoc Networks
• TCP and the Web

11
Example 1(a) TCP over ATM

• TCP throughput on 10 Mbps Ethernet
• approximately 9.5 Mbps
• TCP throughput on 140 Mbps ATM LAN
• approximately 0.2 Mbps!!!
• Why? TCP deadlock problem
• large ATM MTU size, relatively small TCP
  send/receive socket buffer sizes
• interaction between TCP delayed ACKs and Nagles
  Algorithm and socket copy rules
• Solution proper config gt 70 Mbps

12
Example 1(b) TCP over ATM

• TCP throughput and efficiency suffer over ATM
  networks
• Why? The jigsaw-puzzle problem
• TCP packets large ATM cell size small
• one TCP packet many ATM cells (N)
• under overload, ATM switch discards cells
• retransmission at the TCP packet layer
• lose 1 cell, resend N cells for each pkt loss
• Solution
• mark packet boundaries use PPD/EPD

13
Example 2 TCP over Wireless

• Wireless TCP Performance Problems

Low capacity, high error rate
Wired Internet
High capacity, low error rate
Wireless Access
14
Example 2 TCP over Wireless

• Solution wireless-aware TCP (I-TCP, ProxyTCP,
  Snoop-TCP, split connections...)

15
Example 3 TCP over Ad Hoc

• Multi-hop ad hoc networking

Rick
Carey
16
Example 3 TCP over Ad Hoc

• Multi-hop ad hoc networking

Rick
Carey
17
Example 3 TCP over Ad Hoc

• Multi-hop ad hoc networking

Rick
Carey
18
Example 3 TCP over Ad Hoc

• Multi-hop ad hoc networking

Rick
Carey
19
Example 3 TCP over Ad Hoc

• Two interesting problems here
• Dynamic ad hoc routing node movement can disrupt
  the IP routing path at any time, disrupting TCP
  connection yet another way to lose packets!!!
  possible solution Explicit Loss Notification
  (ELN)
• TCP flow control the bursty nature of TCP packet
  transmissions can create contention for the
  shared wireless channel among forwarding nodes
  possible solution rate-based flow control

20
Example 4 TCP and the Web
21
Example 4 TCP and the Web
The classic approach in HTTP/1.0 is to use
one HTTP request per TCP connection, serially.
22
Example 4 TCP and the Web
The persistent HTTP approach can re-use
the same TCP connection for multiple HTTP
transfers, one after another, serially. Amortizes
TCP overhead, but maintains TCP state longer at
server.
23
Our Work CATNIP TCP

• Context-Aware Transport/Network Internet
  Protocol (CATNIP)
• Motivation Like kittens, TCP connections are
  born with their eyes shut - CLW
• Question How much better could TCP perform if it
  knew what it was trying to do (e.g., 14 KB Web
  document transfer)?

24
Motivation for CATNIP TCP

• Main observation
• Not all packet losses are created equal
• Losses early in the transfer have a huge adverse
  impact on the transfer latency
• Losses near the end of the transfer always cost a
  retransmit timeout
• Losses in the middle may or may not hurt,
  depending on congestion window size at the time
  of the loss (because of the TCP fast retransmit
  mechanism)

25
Web/TCP Pain Profile
26
Design of CATNIP

• Make TCP smarter by conveying application-layer
  context to TCP/IP
• modifies socket API

Application
Transport
Network
27
CATNIP TCP Sources

• What could sources do differently?
• Rate-Based Pacing of Last Window (RBPLW)
• Early Congestion Avoidance (ECA)
• Selective Packet Marking (SPM) use a 1-bit field
  in the reserved portion of the TCP/IP header to
  convey packet priority information
• 0 low priority 1 high priority (crucial
  pkts)
• SPM is the most powerful of these
• Implementable using DiffServ codepoints

28
CATNIP in the Internet

• What could IP router do differently?
• If it knew which packets mattered most
• CATNIP-Good avoid discarding them when
  congested (if possible)
• CATNIP-Bad throw them away!

29
Simulation/Emulation Results

• Sources have relatively little control
• IP routers have all the power
• Adding context-awareness at IP routers improves
  both mean and standard deviation of Web page
  transfer times (e.g., by 20-60 for 1-5 packet
  loss)
• SPM and CATNIP-Good provide most of the benefit

30
Summary and Conclusions

• There seem to be performance advantages to
  bending the rules in the layered Internet
  protocol stack
• The general notion of awareness needs to
  explored in lots of contexts
• wireless networks, ad hoc routing, TCP/IP, Web
  caching, mobile computing, adaptive
  applications,
• Many exciting issues to explore!!