Personal Miscellaneous TCP/IP GRID Quality of ServiceMulti-Cast  
Background high-speed tcp Transfer Tests Web 100TCP Tuning  

Web100

Web100 is a project to enable automatic and transparent high throughput transport of TCP/IP. The current implementation (version 1.2 alpha) involves a kernel patch (currently only for version 2.4.16) to allow the ‘live’ kernel level monitoring of TCP variables. Important TCP variables such as the number and type of packets sent and retransmitted, ECN and SACK availability, timeouts and RTO’s, the slow-start threshold (ssthresh) and the congestion window size (cwnd) are all provided on a time-basis, unlike tcpdump/tcptrace which can supply information on a packet-basis. Web100 can also provide an indication of the TCP sliding window like tcpdump/tcptrace, but does not require root access to the system.

At a high level, one could use web100 to simply monitor the instantaneous throughput of a TCP stream and debug simple TCP stack metrics such as sender and receiver window sizes. These values may ‘cap’ transfer rates due to bottlenecks at the sending and receiving hosts rather than any network related bottleneck.

Whilst much of the variable list available through web100 (known as the TCP-MIB) is read-only, a few variables allow write access to allow manipulation of the TCP stack in situ of a TCP connection. These variables currently include adjusting the sender buffer sizes such that a TCP connection can ‘auto-tune’ given the appropriate parameters.

Benefits to the of using such a NMA would include a instantaneous logging facility to any TCP stream, providing a way of characterising and diagnosing all TCP traffic with more than just a throughput/goodput value. Advantages include providing a visual representation of, for example, the effect of different QoS parameters at the host level on a TCP stream.

The introduction of the Web100 group is a good example of the lack of understanding of the interaction of complex transport protocols and the aggregation of ‘self-similar’ [self-similar] traffic. Even though TCP has been around for over 25 years, little is understood about its macroscopic effects. This was most probably due, in part, to the fact that there has never been a need to define a framework for performance measurements with best effort networks. By representing the performance of protocols such as TCP under a methodology that can be easily repeated and compared, we can easily see the benefits and disadvantages of minor or major changes in transport protocol algorithms. Work is currently being conducted by the IETF IPPM [IPPM] group in this respect to define frameworks for network performance measuring and comparisons, and it is hoped that this will lead to a fuller understanding of network protocol interactions and router level implementations and their affect on network performance.

 

Installation Details on how to install Web100 on a linux machine. here.
Variables  
Logvars A script to write out the web100 variables to STDOUT and or a file with customisable trap times and extensive logging. here.

 

Mon, 3 November, 2003 15:14 Previous PageNext Page
 
 
    email me!
© 2001-2003, Yee-Ting Li, email: ytl@hep.ucl.ac.uk, Tel: +44 (0) 20 7679 1376, Fax: +44 (0) 20 7679 7145
Room D14, High Energy Particle Physics, Dept. of Physics & Astronomy, UCL, Gower St, London, WC1E 6BT