Traffic Shaping

1
Presentation by Mark Meiss (mmeiss_at_iu.edu) January
2002
Workstation-Based Traffic Shaping
2
Motivation

• New peer-to-peer file sharing applications emerge
  regularly
• The need for traffic shaping is greater than ever
• How can we limit peer-to-peer traffic without
  starving out other traffic?

3
Problems with Shaping

• Many shapers rely on application signatures for
  shaping
• Often, a signature is just a port number
• Remember learning that port 6699 equals Napster?

4
The Port 80 Effect

• As traffic shaping becomes more widespread,
  peer-to-peer applications will start tunneling
  through TCP port 80
• The traffic we want to shape will become
  indistinguishable from other traffic
• Packet snooping is too difficult and isnt likely
  to help in the long term

5
Our Proposal

• Shape traffic based on different factors
• Apparent amount of demanded bandwidth
• Destination IP address
• Create a notch filter so that users of
  interactive applications suffer the least packet
  loss
• E-mail users shouldnt notice the congestion

6
Our Proposal (cont.)

• Use BGP routes to classify data according to
  destination
• Dont worry about port numbers or packet data at
  all
• Our claim A good PC running an open-source OS
  can shape traffic at least 200,000 packets/sec

7
User-Based Model
User-Based Mathematical Model of Network Activity
8
Sample Run Bandwidth
9
Sample Run Performance
10
Sample Run User Effects
11
Sample Run Zoom In
12
Implementation

• Development platform is Pentium III system with
  64-bit 66MHz PCI slots, running Redhat Linux 7.2
• Software platform is the Click Modular Router
• http//www.pdos.lcs.mit.edu/click/

13
What Next?

• Real-world testing in a few months
• Goal is to get as close to Gigabit line rate as
  possible
• Also working on benchmarking performance of Fast
  Ethernet and Gigabit NICs
• Early results suggest large performance
  differences
• Feedback and suggestions for additional shaping
  parameters are welcome