Workplan proposal task 3.1

Here we describe the steps that have to be taken to meet the objectives that were described in the workplan proposal task 3.1 of the workplan for work package 3.

As the connection between different grid members on both sides of the ocean is of a high speed and long delay nature we will have to be very aware of its special properties. This could lead to the observation that constraints in regard to a possible mismatch between expected and realised transfer capacity exist. Especially packet loss is a very crucial parameter for the throughput obtained with FTP (like) protocol(s). Or: packet loss is disastrous for performance of these handshaking protocols. As we are dealing with telecom equipment on the trans atlantic link (meaning very small buffer room) we can expect problems in places where a high speed connection is fed into a slower one. Substantial effort will have to be put into circumventing the related problems. Mutual feedback with work package 2 will be needed because these properties have direct effect on the performance of higher layer software protocols.

The next steps will have to be taken to come to a precise description of how this long distance link must be monitored:

1. establish the basic properties of the link, e.g. route taken, RTT, one way delay, etc.
2. establish packet loss rate by monitoring closely the equipment involved, for example by letting SNMP generate traps
3. establish packet loss characteristics under different loading conditions, e.g. steady, shaped and bursty traffic.
4. establish the behaviour of the equipment when multiple streams are used as this can cause extensive queueing what again can lead to packet loss.
5. monitoring tools used (which can essentially be the same as in the DataGrid environment) will have to be adapted according to the findings of the above mentioned measurements.
6. come to a recommandation of interface(s) used in the TAG environment as the buffering behaviour can heavily influence the network throughput, e.g. adding burstiness to the output stream.

Results of measurements done can be found on: http://carol.wins.uva.nl/~wsjouw/datatag/lambdaperf.html

The further evaluation of the monitoring tool(s) will be undertaken in task 3.2 and in conjunction with parties in the USA.