On 15.07.14 14:01 , Robert Kisteleki wrote:
... Indeed some anchors are more stable than others. That's not a new phenomenon, I believe it happened to TTM boxes as well. As for what to do with anchors that consistently fail or are down for along time: we don't have a policy defined yet. I'd expect that if a particular anchor proves to be useless for this purpose, then we'll remove it from the pool. (But, after how long? 3-6-12 months, 1 year? We're happy to receive advice from the community about this.)
This discussion shows that RIPE Atlas has reached a scale that makes it necessary to devise quite dynamic ways of "calibrating" measurements. The number of anchors will soon grow such that this is also necessary for anchors. Personally I believe that in the long run it will be necessary to provide calibration data that can be used to interpret the results of measurements and to dynamically change the pools of probes used for specific measurements. For instance, if a probe/anchor suddenly shows an increase in the variance of all its results this may be a network phenomenon or a measurement artefact. In any case it *may* be reason to disregard or scale the results when using them in specific measurements and studies. I am currently working on this. This is slow work because it is not straightforward and also because I have had some unexpected other duties in the governance area. Daniel