[atlas] Flow-id of built-in traceroute measurement

Hi Philip,

Thank you for replying.

I have some observations, but not really statistics, as they only tell the story of my fairly limited sample set.
I selected built-in traceroute measurement to b-root from 128 probes hosted in European DC.
123 of them have no more than 16 different IP paths over a weeks' time, i.e. 336 traceroute measurements in all.
3 of them have more than 50 different IP paths, among which 1 have more than 200.

Regards,
wenqin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ip_diver_cdf.pdf
Type: application/pdf
Size: 5149 bytes
Desc: not available
URL: </ripe/mail/archives/ripe-atlas/attachments/20160408/b89294df/attachment.pdf>
-------------- next part --------------

> On 06 Apr 2016, at 10:36, Philip Homburg <philip.homburg at ripe.net> wrote:
> 
> Hi,
> 
> On 2016/04/05 20:53 , Wenqin SHAO wrote:
>> I have some questions concerning the flow-id of built-in traceroute measurements and are seeking for your help.
>> 
>> As many of you may have already noticed, the flow-id of the built-in traceroute measurement periodically varies form 0 to 15, in a circular manner.
>> With this behaviour, it is supposed to reveal the diversity of underlying IP paths.
>> 
>> In the case that there are more than 16 unique IP paths between a source and destination pair, could this scheme reveal all of these IP paths?
>> If it can, which leads to more than one IP paths per flow-id? Is it normal, from a load balancer point of view?
>> If it can not, then why not a wider range of flow-id (more nuances in header hash results), say [0, 63]? Is it related to common realisation of load balancer?
> 
> Basically, the wider the range, the longer you have to wait for two
> traceroute results that take the same path.
> 
> Of course, if you want to investigate one particular path, you can just
> schedule your own measurement with different parameters.
> 
> I have not seen any analysis that estimates what percentage of source,
> destinations pairs would have more than 16 different paths. If that
> percentage is significant, then it's worth considering raising the value.
> 
> Philip
>