[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[ih] Why did congestion happen at all? Re: why did CC happen at all?

On Aug 31, 2014, at 1:52 PM, Detlef Bosau <detlef.bosau at web.de> wrote:

> So there must be something in our networking world, which is different
> from that real world, and therefore, I ask questions.

Are we ready to talk about the real world?

In the real world, the law of large numbers dominates everything else. Core routers are now pushing 100Gbps per interface and carrying millions of flows in parallel. Real world networks are engineered with substantial amounts of over-provisioning, and not unreasonable amounts of buffering. Packet drops are still (relatively) very rare. Queueing delay does occur and is quite sufficient to trigger Slow Start which pretty clearly is effective at reducing congestion.

For all practical purposes, it works just fine. 

Bursts of traffic still happen. If they?re very bad, they cause queueing. If they?re horrible, they cause drops. If you?re concerned about this, the thing to look at is how to avoid a burst that comes from aggregating a very large number of pseudo-random sources with serialization during the aggregation. Short of the overhead of global admission control, and its concomitant exorbitant overhead, there?s not an obvious architectural path forward.