Implications for QoS provisioning based on traceroute measurements

1
Implications for QoS provisioning based on
traceroute measurements
Milena Janic, Fernando Kuipers, Xiaoming Zhou and
Piet Van Mieghem Delft University of Technology,
The Netherlands QofIS 2002 October 16th
2
Overview

• The goal and the method
• The results of the analysis
• Implications for Future Internet Services
• Implications for e2e QoS provisioning
• Implications for multicast deployment
• Conclusions

3
The goal and the method

• The goal to provide insight into provisioning of
  Future Internet Services, based on RIPE NCC
  traceroute measurement data.
• The method
• Errors in traceroute analyzed and categorized.
• A graph approximating a part of Internet created
• Polynomially distributed link weights assigned

4
RIPE measurement configuration
The traceroute data provided by RIPE NCC (the
Network Coordination Centre of the Réseaux IP
Européen) in the period 1998-2001.
5
Constructing G1
the most frequently occurring non-erroneous path
for s-d pair (A,B)

• 386 most frequently occurring non-erroneous
  paths
• 1888 nodes and 2628 edges.

• G1 represents interface map not router-level map!

6
The topological properties of G1
Uniform recursive tree
Any new node N has equal probability to be
attached to any of the N-1 nodes already in the
tree.
7
The hopcount in G1
8
The pdf of node degree in G1
9
The agreement of G1 and URT?

• The intersection of the trees computed
• (a) the node number in the intersection small
• (b) the common nodes increased the degree by 1
• The discrepancy with the result of Faloutsos
• interface map instead of router-level map
• small number of test boxes

10
The analysis of l.w. structure
11
The analysis of l.w. structure (contd.)
?hhRIPE-hSIM
12
The dominance of Internet paths

• IP packets follow different routes in 3 years
  period!
• Possible causes growth of the Internet, load
  balancing, multi-homing, changing SLAs, failures

13
Implications for e2e QoS provisioning

• Alarming situation if this behaviour is caused by
  failures
• Important to have a smart update strategy and
  to structure Internet into more hierarchies

14
Implications for Multicast Deployment

• The netto gain (savings minus costs) must be
  larger than the netto gain for unicast

• The multicast efficiency/gain (Van Mieghem et al)

• The average hopcount in r.g.

• Unicast uses on average fN(m)mEhN links ?
• the ratio gN(m)/ fN(m) -a good estimate for
  efficiency!

15
Implications for Multicast Deployment (contd.)

16
Conclusions

• A strong resemblance of the properties of graph
  G1 to those of the uniform recursive tree has
  been found.
• This might enable network operators to compute
  the multicast efficiency over unicast
• No clear conclusions on the link weight structure
  of the Internet could be deduced.
• The Internet has to be made more robust

17
Questions?