On 8 Jun 2012, at 09:47, "Kerry Milestone" <[email protected]> wrote:
> I have seen a transit provider who is/was preferring small packets > (traditional VoIP) but once the packets reached a larger size such as > containing the HD and extra frame rate components these were going down a > larger (layer 2, different framing etc) bandwidth channel with some extra > internal hops. However it increased latency from the base layer such that > there was too much to actually be useful to build on the baselayer stream > presentation. The only (ok.. easiest) fix in this case, was to move to a > different peer who actually wasn't treating VoIP with any difference. i wonder if thats what really happened ? it seems incredible that packet size would dictate a preferred path. more like the VoIP traffic being classified and the HD VoIP not being classified in the same way. where you able to test using a different application? > Now for a telco to know/detect and oblige priority for particular streams > would be tricky. now that's an understatement! but we do it with DPI for peer to peer and other traffic so it can be done. I would argue that it's an expensive solution - probably more than the cost of the bandwidth and certainly not with the margins on voice (!) and not something that in my view aligns with a net neutrality direction. > > On 07/06/12 13:35, Matthew Ford wrote: > > > > http://www.iab.org/cc-workshop/ > > > > Mat > > > > On 07/06/12 19:49, Neil J. McRae wrote: >> in some cases yes - ATM based DSL lines are particularly bad at coping >> with VoIP under congestion - I presented some stats from my days at COLT >> on this at UKNOF4 I think. >> >> However even with QoS and other priority chaos there is still no >> guarantee even on IP friendly access links. > > On 7 Jun 2012, at 13:12, Justin Finkelstein wrote: > > > Good point - I hadn't thought of that. If we consider streaming video > > (within the context of communications) we're then talking about 2-way > > live video feeds. I imagine at this point, the bandwidth requirements > > would jump rather considerably and the ability to shape/prioritise > > this traffic would become more complex as (IIRC) we're not just > > talking about small UDP packets any more. > > > > > -- > The Wellcome Trust Sanger Institute is operated by Genome Research Limited, a > charity registered in England with number 1021457 and a company registered in > England with number 2742969, whose registered office is 215 Euston Road, > London, NW1 2BE.
