|
Description |
The Used Metrics |
Authors
The network topology gauge was meant to measure various metrics over WAN.
It provides a general purpose library that can be used by any topology
and locality aware application, as well as a the gauge application itself,
topologyd (stands for topology daemon), that was run successfully in the
Planet-lab environment.
The network topology gauge is composed of three major components:
- Measuring worker
- This worker receives a list of the other hosts, and one by one
it performs all the measurements described above. It continues to
do so until the daemon is shut down.
- Measuring server
- Some of the measurements (e.g. ping over TCP, upload/download
bandwidth) require cooperation from the measured host. The server
listens on a known port, and handles requests from the measuring
hosts. Each request is given to a newly created thread and handled
there. Each request begins with a unique byte indicating its type
to allow the server peek the correct handler.
- Monitoring Web Server
- Developing distributed software has never been an easy task.
Monitoring approximately 200 nodes around the globe seems like a
administrative nightmare. In order to simplify this task a small
web server was embedded in the topologyd, allowing the user to
quickly view the status of the server, the measured values and
retrieve the accumulated logs. In addition, the results can be
shown in a visual manner, where other hosts are displayed as
flags on a map. Three maps are available: global, United States
and Europe.
In order to ease changing the monitored node, each reference
to another node is a link to that node's web monitor. Reference
can be its entry in the distance table, or its flag in the map
view. These links enable the user to quickly jump between nodes
and see the current system status from different views.
Screenshot of the monitor can be seen
here. Each flag represent a
measured host, in its actual geographical location. The flag's
color indicates the distance of that host. When the mouse pointer
hovers over the flag, a small tool-tip open with various
information on the host such as DNS name, IP, institute and
geographical location is shown.
The user can change the weights of the metrics, so it can
emphasize the relavant metrics. If the weight is set to 0
(zero), The corresponding measurement is not performed.
The code was compiled and tested on the following systems:
- RedHat 9 / Planet-Lab v2
- Fedora Core 2
- Libranet (Debian)
- These metrics use raw sockets to send ICMP packets.
In Planet-Lab no root access is needed, as it provides the
safe raw sockets mechanism.
- The number of autonomous systems metric is based on
the results of the number of hops metric.
The original authors are Danny Bickson
and David Rabinowitz.
For questions and support you are welcome to send mail to
daniel51 @ cs.huji.ac.il.
|
