Measurements
From FIpedia
The Disagreement
The role of measurements and the ability to measure seems to be the point of contention, i.e.,
- What needs to be measured?
- Is the ability to measure crucial for finding new designs?
- Is the ability to measure crucial for evolving current designs?
Viewpoint 1: Measurement is not a critical input to the design of future network architecture
I will somewhat overstate this contrarian position to be provocative. Measurement of the existing Internet is very valuable for several reasons, e.g. to find things that are broken, to understand current patterns of usage, and perhaps to predict future patterns of usage. Measurement will always be necessary for debugging and as a part of informing the larger social and economic debate around the use of the Internet.
But, as we have learned from the past, patterns of usage change rapidly as the capabilities of the network change, and as the creativity of application designers trigger new patterns of usage (e.g. P2 sharing). If our goal is to design a new network, we should start by discussing the range of usage patterns we want to support, and then build a network that supports them. Measuring what folks are doing today is not a good way to define our ideal tomorrow.
When I started working on systems, Jerry Saltzer (my PhD thesis advisor) said to me: "You cannot decide where to build a bridge by counting the people swimming the river".
Viewpoint2: There exists many traffic measurement projects
QoS in Future Internet depends strongly on the ability of network operators to cope with customers demands in terms of bandwidth (BW) and parameters like jitter, delay or bit error rate (BER), that can be specified in the Service Level Agreement (SLA).
Several projects have been developed to deal with IP traffic measurement: MOME ([1]), LOBSTER ([2]), ETOMIC ([3]),DIMES ([4]), RIPE ([5]) or the set of efforts federated in the Cooperative Association for Internet Data Analysis ([6]). They include their own infrastructures or simply store data or provide the means to perform traffic measurements; some general purpose infrastructures are also avilable throughout the world: GEANT2 ([7]), Perfsonar ([8]), and new ones are being deployed so as to be used for Mo/Me techniques testing.
These are data enough to realize the importance of Mo/Me technologies in FI development for research groups, network operators, service providers and even governments (or Internet governance institutions): Cf my argument in[[9]].
Perhaps the diversity of Mo/Me initiatives migth become a serious drawback when people not directly involved in their development want to use them. They use different techniques and provide data in different formats. Recently MOMENT project ([10]) has been designed to serve as an universal mediator to those different Mo/Me infrastuctures by means of a semantic interface.
Moderator
Dirk Trossen
Leave a Comment
