[c-nsp] Maximum traffic on Gigabit Ethernet

[Paul Cosgrove]

Hi Nick,

I think you may have an error there.  Your calculations suggest that traffic
in on a 1Gbps interface will fill a 1.3M Byte receive buffer in a little
under a 1/10 of a second.  While the preamble and inter frame gap will slow
the buffer fill rate (the degree depending on the packet size), that seems
very slow.

Presumably the buffer would also be processed by the router, and that would
need to be accounted for over a 5 min interval.

Paul.

On Mon, Oct 3, 2011 at 4:37 PM, Nick Hilliard <nick at foobar.org> wrote:

> On 03/10/2011 14:47, Manaf Al Oqlah wrote:
> > What is the maximum traffic that a gigabit Ethernet interface can handle
> > on Cisco 7600 router RSP720-3C-GE before dropping packets . we are able
> > to reach 750 mbps / 125 mbps input/out rate!
>
> The maximum traffic that a gigabit Ethernet interface can handle before
> dropping packets on any interface capable of handling line rate
> transmission is exactly 1,000,000,000 bits per second - no more, no less.
>
> I think you're confusing "maximum traffic" with "maximum traffic averaged
> out over a period of X seconds, where X is larger than the
> deterministically longest buffer time available on that specific port".
>
> To understand why your question is meaningless, consider the situation
> where a GE port receives exactly 2gbit/sec for one second, then spends the
> next 299 seconds with no traffic at all.
>
> Because the port could only handle 1Gbit/sec out of the 2Gbit/sec received,
> this will register as a total of 1Gbit/sec of traffic over 300 seconds,
> i.e. 3.333 mbit/sec.  However you will see 50% packet loss, because the
> port could only handle half the traffic it expected to process.  Actually,
> it won't be quite 50% due to packet buffering, but you get the idea...
>
> So from a mathematical point of view, let's say you're using a
> WS-6748-GE-TX card which has 1.3M of buffer space per port and that you're
> using 5 minute load counters.  This works out as 96ms worth of traffic.  So
> you are guaranteed that regardless of the input traffic load, the first
> 96ms worth of traffic received on the port will not have any packet drops -
> note that this is not the same as the first 96ms worth of traffic
> transmitted to the port.  If you figure out the maths, it means that
> without prior knowledge of the traffic profile, the maximum 5 minute load
> average on that GE port where you are _guaranteed_ not to have packet drops
> is 320kbit/sec.  Freaky, huh?
>
> This is nit-picking though.  In real life, the number is highly variable,
> and usually ranges from about 650mbit/sec upwards, depending on traffic
> type.  imix traffic is usually much more forgiving than bursty lan traffic.
>
> Nick
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>