Passive Capacity Estimation: Comparison of Existing Tools

1
Passive Capacity EstimationComparison of
Existing Tools
Taoufik En-Najjary Guillaume Urvoy-Keller 18 June
2008
2
Definition

• Capacity The capacity of a path is defined as
  the maximum IP-layer throughput that a flow can
  get on the path, without any cross traffic.
• Available Bandwidth The maximum rate at which a
  new flow can send without impacting the rate
  achieved by the existing flows.

3
Active Tools

• Active techniques
• inject a probe packets in the network
• Control over the measurement
• Not suitable for large scale measurement
• Probe overhead !
• Need to access to both the sender and the
  receiver
• Cannot be applied to the large set of Internet
  traces

4
Passive Tools

• Extracts the capacities from TCP traces
• TCP is the most popular transport protocol (gt90)
• TCP sender often injects packets in pair in the
  network delayed acks strategy
• Work with measurement collected any where in the
  path.
• But depend on the available data
• Need a minimum number of packets per connection

5
Passive Capacity Estimation

• Based on packet pair dispersion techniques
• Path P a sequence of links from source S to
  destination D
• Two packets of the same size L are sent
  back-to-back from S to D along a path P with link
  capacities
• Dispersion at link i is the time interval
  between the transmission of 1st bytes of the 1st
  packets and the 1st bytes of the 2nd packets
• Capacity estimate
• If no cross traffic, every bandwidth estimate is
  equal to the capacity

6
Passive Capacity Estimation

• Histogram Extracted fromWhat do Packet
  Dispersion Techniques Measure?', Dovrolis et al,
  Infocom 2001

The capacity distribution is multimodal which
mode is the capacity mode?
7
Comparison of passive Tools

• Many tools
• PPrate, MultiQ and Nettimer
• How accurate?
• How many samples they need to return an estimate?
  
• Comparaison on PlanetLab
• Comparison with an active tool Pathrate
• Consistency The effect of the number of samples
• Comparaison on a large ADSL traces
• CDF
• Scatter plot
• Histograms

8
Planetlab Experiments

• Selection of 33 Paths between Planetlab nodes
• On each path
• SCP transfer of a 20 Mbytes file
• Collection of tcpdump traces at the receiver side
• Run Pathrate on the same path
• Repeat procedure 10 times

9
How accurate?

• Bandwidth reported by PPrate (o) and Pathrate ()

Passive and active tools have the same behaviour
10
How many samples?

• 20 MB ? 12000 data packet
• PPrate return the same estimate with 300 packets
• Nettimer and MultiQ need more than 5000 packets
  to return a consistent estimates

11
ADSL traces analysis

• Public ADSL traces http//m2c-a.cs.utwente.nl/re
  pository
• Traffic collected close to ADSL clients
• The only information we have
• The access link capacities range from 256 kbits/s
  to 8 Mbits/s
• We apply the three tools on the ack stream and
  data stream of HTTP connections
• We consider only the streams with more than 300
  samples
• 8 of HTTP connections

12
Estimated Capacitie of HTTP servers

• Peaks at 10 and 100 in accordance with Ethernet
  network
• Nettimer under estimates the servers capacity

13
Estimated Capacitie of HTTP servers

• Tools are not agree
• Need to look to capacity histograms

14
Estimated Capacitie of HTTP servers
PPrate 1 Mbps, MultiQ 100Mbps
PPrate 50 Mbps, MultiQ 400Mbps
15
Estimated Capacitie of ADSL Clients

• Nettimer Under estimates the capacity of ADSL
  clients

16
Conclusion

• Nettimer do not work on acks stream, and often
  under estimates the path capacity
• MultiQ sometimes overestimates path capacity
• There is a number of cases where it is difficult
  to determine which tool is actually right
• PlanetLab is not sufficient to assess the
  accuracy of measurement tools

17
Thanks !