[Battlemesh] The Flent testing tool
david at lang.hm
Mon Aug 17 01:42:12 CEST 2015
On Sun, 16 Aug 2015, Jonathan Morton wrote:
>> On 16 Aug, 2015, at 20:13, Henning Rogge <hrogge at gmail.com> wrote:
>> I think Jonathan, the author of Cake said that it shouldn't be that difficult to convert the rest of Cake to "time" instead of bytes…
> More precisely, I would have to look carefully to *find* calculations which are still essentially bytes-based rather than time-based.
> There are subtleties, though. Many parts of Cake assume that the drain rate is roughly constant (through the shaper), and use bytes with a pre-calculated scale factor (or limit) as a proxy for time. A version of Cake specialised for wifi would require all those assumptions to be revisited, because:
> - Drain rates vary over time, depending on signal quality and contention.
> - Drain rates are station-specific, not global to the interface.
> This is still entirely doable; Cake is not a huge body of code, and I suspect
> that moving it into the wifi stack would entail a considerable rewrite in any
> case, to fit a wifi-specific API. We may end up calling it something other
> than Cake after such a rewrite, leaving Cake itself as a conventional qdisc
> for wired links (as it is now).
It also sounds like we need a noop qdisc in the main stack. this would just pass
things to a secondary set of queues (the wifi specific queue, or potentially
even the offload hardware type of stuff)
More information about the Battlemesh