Tbwtools

From GEANT2-JRA1 Wiki

1 Tbwtools - Debugging TCP Performance

Tbwtools - Debugging TCP Performance

Description

Tbwtools is a set of tools and a framework for integrated debugging of TCP performance using information obtained from several sources. Tbwtools use a combination of active and passive monitoring and are installed entirely on end hosts, no access to network infrastructure (routers, etc.) is required.

Features:

shows characteristics obtained from tcpdump tracefile (e.g., current throughput or frequency of segment retransmissions)
shows TCP runtime variables obtained from web100 kernel instrumentation (e.g., current TCP state or smoothed RTT)
show socket options obtained from bulk utility (e.g., current congestion window and ssthresh)
user does not need anything else than a standard web browser with Java support
the framework allows to make a test connection between users PC and a remote server or between two remote servers (running tbwMP)
available measure points (tbwMPs) can be obtained from Lookup Service
tbwtools pass through NAT, which can be between user interface in tbwApplet and tbwMPs
user interface in tbwApplet is not mandatory, you can also send XML requests directly to tbwMPs, but of course user interface is convenient

For more detailed description, see CESNET technical report 6/2006

[edit]

Measurements

User interface for tbwMPs installed in CESNET is accessible at https://lego.cesnet.cz/tbw (requires username and password, see here) and only works if your PC has IP address in CESNET network, because access to LS and tbwMPs in CESNET is restricted. If you want to test performance to tbwMPs in CESNET, send your IP address to contact below

[edit]

Installation

Download the latest version of tbwtools (requires username and password, see here). Read INSTALL file the there.

[edit]

Discussion

What improvements to tbwtools do you suggest?

Identified shortcomings (SU):

Produced graphs require expert knowledge of TCP operation to understand them and make conclusion or hypothesis about causes of performance limitation.
The list of ticks to select graphs to be produced should be preset to some practical set of graphs. Users should be able to store their own profiles (sets) of requested graphs.
The extension of the previous point is that tbwtools should allow semi-automatic performance debugging in the following style:
- initial set of most useful graphs is automatically produced and comments and suggestions to the user are given, such as "if graph A has lots of fluctuations, it means ..."
- based on suggestions in the previous step, user should be able to easily select one of several branches for further investigation of the problem (we have narrowed the cause of the problem)
- user clicks on one of several buttons, a new set of graphs or other information is produced and the three steps are repeated until performance problem is identified

Is anybody willing to contribute to such semi-automatic performance analysis?

[edit]