On 20:12:45 Nov 16, Stuart Henderson wrote: > > Say you have a 10Mb ethernet feed, plugged into an > unmanaged switch with a bunch of other people in the > building connecting to other ports, who sometimes use > up all available bandwidth on the uplink, and other > times use nothing.
I am not sure if this situation can arise in the present scenario. Most ADSL links are used at home with hardly more than one user at one point of time.I was referring to a case of congestion at the upstream router. If that accounts for the variation OP is seeing then measurement can do something. Usually downlink paths tend to get congested depending upon who is downloading how much but the uplink bandwidth is mostly unused. Moreover the uplink capacity actually widens quite a bit the moment the packet crosses the ADSL link and reaches the exchange. However in case your point applies to this situation then surely any amount of statistical analysis won't help. > > Now you want to take whatever of that uplink is available > to you, and share it fairly between users, giving priority > to some over others. > > Obviously if you set a queue at 10Mb you'll have problems > sometimes. But if you set it at the "average", you'll Obviously "average" will not help here. > A) miss out on a lot of bandwidth most of the time and > B) still have problems when the connection is heavily > used by people in the building who aren't downstream of > your PF box. > > Similar sort of deal with a normal shared-access satellite > system. > > Someone please correct me if I'm wrong, but I think that > congestion is defined as "bw wanted > bw configured on the > interface in the 'altq on' definition". Problem there is > you can't tell what is available at a given time. > > From what you quoted Jonathan: > > "Queues with a higher priority are preferred during congestion > over queues with a lower priority as long as both queues share > the same parent" > > OpenBSD's pf.conf(5) fits a little more information into about > the same space: > > "Priq queues with a higher priority are always served first. > Cbq and Hfsc queues with a higher priority are preferred in > the case of overload." > > I don't think it's possible to do exactly what's wanted with > the existing altq disciplines. Priq would starve out lower > priority queues; cbq/hfsc would have the problem that they > can't identify an overload on this sort of uplink. Can someone shed some light on this please? I for some reason cannot imagine that Daniel or anyone would have missed this point.. regards, Girish
