deTector is a real-time, low-overhead and high-accuracy monitoring system for large-scale data center networks. At its core is a carefully designed probe matrix, constructed by a greedy path selection algorithm that is scalable and probe overhead minimizing.
Python 2.7.6.
The system configuration file config.xml in the Controller specifies all details such as Controller/Diagnoser IP, listening port and ping interval. To run deTector on a cluster, you need to update this file according to your cluster configuration.
You need to start Controller first using the following command:
$ python controller.py
The Controller will read the configuration, compute the probe matrix and start a HTTP server.
To start the Diagnoser, run:
$ python diagnoser.py controller_ip controller_port
The Diagnoser needs to connect to the controller to get the probe matrix.
Next, start Responders on all servers using the command:
$ python responder.py controller_ip controller_port
Finally, run Pingers on servers who ping others:
$ python pinger.py host_ip controller_ip controller_port
For ease of deployment, you can direct run script.py to deploy deTector in a cluster after you finish basic configuration and configure auto-login ssh. This script reads the configuration from config.xml and by default, it runs Controller and Diagnoser on localhost.
To do loss localization, you also need to start failanalyzer on the same machine with Diagnoser using the command:
$ python failanalyzer.py
Note that deTector requires source routing in the cluster to controll packet path. Packet encapsulation and decapsulation (e.g., IP-in-IP) is a general solution and should be supported on commodity switches.
Please read Architecture.md and Workflow.md for more system design details.
Read the technical report for the core algorithm of deTector.