Hi there, On Jan 11, 2013, at 05:16 , Brian J. Murrell wrote:
> On 13-01-11 07:29 AM, etienne.champet...@free.fr wrote: >> Hi >> >> http://wiki.openwrt.org/doc/howto/packet.scheduler/packet.scheduler > > Note that OP described wanting to distribute "download" bandwidth for > users. The document you posted above is about shaping "upload" bandwidth. > > As the above document states, one cannot really "shape" download > bandwidth but can only "police" it by dropping packets as they come in, > and then, that only works for TCP and not UDP. I wonder about whether that is actually right. Looking at the definitions of shaping and policing it is quite clear that in the real world any shaper will also police (and if only by virtue of limited buffers at one point packets will be dropped, turning a shaper into a policer). Adding a shaper to san ingress policer offers the following advantage: the developing queue, especially if the queueing is per flow or close to per flow (as in a number of different queueing disciplines available) will give the policing component more information to pick which packets to fro or mark. Thus in the real world shaping can be considered, as a fancy way of policing, so any shaper will fall back being a policer (worst case scenario all buffers are filled and a new packet arrives). Neither shapeing nor only policeing will be able to help against an inelastic UDP flood on the downstream. Especially if said flood consumes all link's bandwidth so even if the receiver drops all UDP packages no real traffic will make it through the bottleneck. This can only be solved on the head-end of the downstream; but inelastic UDP traffic is quite uncommon outside of DOS attacks, so this has nothing to do with the question of to shape or not to shape. Where shaping will help is if several (longer-running) TCP flows compete the downlink, then shaping can help to distribute the speeds of these individual flows to "fairly" share the downstream bandwidth (depending on your shaper setup and definition of fair). All of these facts (if actually) true seem to make the argument for using a shaper not a policer even on the remote end of a "bottle-neck" link like a typical cable/dsl/lte connection. Now, I could be just full of it, so please show me where my reasoning is off, and why policing is better than shaping. best Sebastian > > b. > > > _______________________________________________ > openwrt-devel mailing list > openwrt-devel@lists.openwrt.org > https://lists.openwrt.org/mailman/listinfo/openwrt-devel _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
