Network Diagnosis

1
Network Diagnosis

• Reshma Khatri,Pallavi Phene
• reshma, phene_at_cs.uh.edu
• 11th May 2003

2
Outline

• Introduction
• Background
• Methodology
• Results
• Future work
• Conclusions
• References

3
Introduction

• Motivation
• Performance of distributed applications benefits
  from selection of paths which will maximize
  communication speed.
• Thorough network analysis required
• Goal
• Develop a methodology for predicting the best
  paths for the application

4
Background

• Network Topology
• Intrusive Methods
• Sub-network Independent Methods

5
Background (contd)

• End-to-End Throughput
• Amount of data that may be transmitted over a
  path during an interval of time
• Capacity
• Available Bandwidth
• Bulk Transfer Capacity (BTC)
• Techniques
• ttcp
• TCP-level methods
• Application-layer methods

6
(No Transcript)
7
Background (contd)

• Bottleneck Links
• One-Packet Techniques
• Predicts the time needed for one packet to
  travel across L-1 links before the Lth node
• Each link contributes to some transmission delay
  and latency

8
Background (contd)

• Packet Pair Techniques
• Measure the smallest (bottlenecked) bandwidth
  link along a path
• Predict the difference in arrival times of two
  packets traveling from the same source to the
  same destination
• Basis - packets sent close enough together in
  time to cause the packets to queue together at
  the bottleneck link

9
(No Transcript)
10
Background (contd)

• Multi-Packet Techniques
• Take latency into account along with all the
  packets in a single flow

11
Background (contd)

• Packet Loss
• condition of a packet apparently being sent from
  one end of a path, but never being received at
  the other
• causes
• corruption of packets due to unfavorable network
  conditions
• intentional dropping of packets at routers due to
  network congestion

12
Background (contd)

• Techniques for detecting packet loss
• TCP netstat
• UDP transmitter - receiver technique
• ICMP ping, traceroute

13
Background (contd)

• Network Delay
• Time taken from node-to-node in a network
• Network Jitter
• Variation in delay over time from node-to-node
• Tolerable jitter on the network is affected by
  the depth of the jitter buffer

14
Methodology

• Design Overview
• 2 Components
• Topology Identifier
• Path Predictor

15
Discover Network Topology

• Periodic evaluation
• Begin with a topological model consisting of all
  known nodes in the network.
• At each iteration of the process
• Determine path between each source-destination
  node pair add any newly discovered links or
  nodes to model
• Detect external gateways for every known node and
  add to model
• Perform Source Routing to detect nodes and links
  in model which are no longer reachable

16
Determine End-to-End Throughput

• For every source-destination pair, send as many
  packets as possible down every path for a fixed
  duration
• Fixed period should be at least 5 times the
  round-trip time for that path

17
Detect Potentially Bottlenecked Paths and Links

• Perform packet-pair technique between every path
  for each source-destination pair
• Identify potentially bottlenecked paths based on
  application requirements
• Perform packet-pair technique on each consecutive
  pair of nodes along each potentially bottlenecked
  path
• Identify potential bottleneck links based on
  application requirements

18
Recommend Best Path

• For a source-destination pair, select paths with
  highest throughputs
• From those, select the one with the fewest
  bottlenecks

19
Setup

• Nodes
• redhat01.cs.uh.edu, redhat01.cs.uh.edu,
  themis.cs.uh.edu, creek.uh.edu, and
  129-7-236-80.dhcp.uh.edu
• Implementation
• traceroute, route
• Perl scripts

20
Results Network Topology
21
Results - Throughput
22
Results Throughput (contd)
23
Results Potential Bottleneck Bandwidths
24
Future Work

• Selection of best set of execution nodes for the
  distributed application by mapping an Application
  Graph to a Resource Graph

25
Resource Graph
Application Graph
26
Conclusions

• Analysis of a network is a complex task
• Numerous measurements possible
• Comprehensive analysis of the current network
  attributes is not feasible, so applications must
  prioritize attributes based on their requirements
• Preliminary results from a small-scale testing of
  this model have been obtained, and future work
  will involve expanding this to a large-scale
  implementation

27
References

• M. Allman, Measuring End-to-End Bulk Transfer
  Capacity, Proceedings of the ACM SIGCOMM
  Internet Measurement Workshop 2001.
•  
• M. S. Borella, D. Swider, S. Uludag, and G. B.
  Brewster, "Internet Packet Loss Measurement and
  Implications for End-to-End QoS", In Proceedings
  of the International Conference on Parallel
  Processing 1998.
•  
• E. Celebi, MS Thesis, Performance Evaluation of
  Wireless Mobile Ad Hoc Network Routing
  Protocols, 2001. http//www.cmpe.boun.edu.tr/ emr
  e/research/msthesis/
•  
• M. J. Coates, R. Castro, M. Gadhiok, R. King, Y.
  Tsang and R. Nowak, Maximum Likelihood Network
  Topology Identification from Edge-Based Unicast
  Measurements, Proc. ACM Sigmetrics, Marina Del
  Rey, California, Jun. 2002.
•  
• M. Jain and C. Dovrolis, "End-to-end available
  bandwidth Measurement methodology, dynamics, and
  relation with TCP throughput," in Proceedings of
  ACM SIGCOMM '02, Pittsburgh, PA, Aug. 2002.
•  
• K. Lai and M.Baker, "Measuring Link Bandwidths
  Using a Deterministic Model of Packet Delay," In
  Proceedings ACM SIGCOMM00, Stockholm, Sweden,
  September 2000.

28
References(contd)

• K. Lai, M. Baker, Nettimer a tool for measuring
  bottleneck link bandwidth, Proceedings of the
  third USENIX Symposium on Internet Technologies
  and Systems, San Francisco, CA, March 2001.
•  
• R. Wolski, "Dynamically Forecasting Network
  Performance Using the Network Weather Service",
  Journal of Cluster Computing Volume 1, pp.
  119-132, 1998
•  
• D. C. M. Wood, S. S. Coleman, and M. F. Schwartz,
  Fremont A system for discovering network
  characteristics and problems, In Proceedings of
  the USENIX Winder Conference, pages 335-348,
  January 1993.
•  
• Cable Modem Troubleshooting Tips Packet Loss,
  http//homepage.ntlworld.com/robin.d.h.walker/cmti
  ps/loss.html
•  
• Network Performance Testing, http//www.netcordi
  a.com/tnm/tnm31/ttcp.htm