Transcript dimes-etomic-june-2006

Packet train experiments in
DIMES and ETOMIC
Ohad Serfaty
Joszef Steger
May 2006
Preview
• PacketTrain Measurements
• Results and Difficulties
• Solutions
• Future work
PacketTrain measurement
scheme
UDP Packets
Port 7777
Results are
compressed and
uploaded
Etomic Agents
Dimes Agents
Dimes Server
Measurement automation
• Agents receive a script with Traceroutes
•
•
•
•
commands and PacketTrain commands
Etomic Agents receive a script with a listen
command
At the end of the listening period – The Etomic
agents compress the results and send it to DIMES
Results are inserted into the databases , spanning
trees are created
Tomography
PacketTrain experiment design
• Agents are picked , one from each user
• Etomic nodes are picked
• Each agent sends 10000 packet trains , in
intervals of 50ms-100ms
PacketTrain experiments
• We conducted 4 full experiments
• 2 Partial experiments
• 10/5/2006 • 11/5/2006
• 140 Dimes Agents , 4 Etomic nodes
• 2 Large scale experiments
• 28/5/2006
• 30/5/2006
• 241 Dimes Agents , 9 Etomic nodes
Results - Agent 4827
Results - Agent 4827
Spanning tree – Agent 5245
Messy Time series
Messy Time series
Messy Spanning Tree
Geographic span
Tomography requirements
• Well behaved time series
• Well behaved Traceroute spanning tree
• Exact match between Agent and IP
Analysis obstacles
• Indefinite slope in time series
• Too many packets from the same IP
• Traceroute data is corrupt
• No accurate matching between Agent and
IP address
• Solution : Traceroute back to the client
Java Network Stack
•
•
•
•
•
Encapsulated Packet send / receive
Cross platform : windows , Linux , Mac
and Etomic
Java - Write once run anywhere
Will be the base of future DIMES
measurements
Traceroute back to the Agent
IP to agent matching
IP to agent matching
IP to agent matching
IP to agent matching
Future work
• Add Agent id to Packet train module
• Tomography analysis
• Install a measurement infrastructure
similar to DIMES on colbud cluster