🚀 go-pugleaf

dave@pmafire.inel.gov (Dave Remien) writes:

> In article <1991Sep1.195029.16374@aber.ac.uk> pcg@aber.ac.uk (Piercarlo Grandi) writes:
> >Probably what you mean is that you will have a separate Ethernet cable per
> >computer. This of course means that each machine would have its own 0.8MB/s
> >Ethernet dedicated to itself. The problem is that I doubt very much that
> >this would buy you anything at all, as I would expect the CISCO router not
> >to be able to sustain more than 0.8MB/s aggregate anyhow, as routers are
> >normally not designed to sustain the maximum combined bandwidth of all the
> >Ethernets they link. Does anybody know for sure what is the maximum thruput
> >of a CISCO router?

> Ours forwards at about 15000 packets per second, and routes at about
> 2000 packet/sec. The newer ones are supposed to about double that.

Amazing facts #2! Shannon is in a jolly good mood if he gives so many
dispensations these days:

> Lessee ---- 15000 * 1500 = 22.5MBytes/sec, a tad faster than Ethernet
> will go.

25 times the bandwidth of a single Ethernet. Impressive! CISCO must be very
generous to give customers so much extra capacity, as their routers rarely
connect more than 2-3 Ethernet segments, and they overdesign their products
to handle ten times the maximum load on their average configuration.

> Even routing implies (assuming max size legitimate packets) 3MBytes/sec.

Here I detect the possibly improbable assumption that it can forward or
route maximal size packets at the maximum rate, which seems incredibly
optimistic, because 22MB/s aggregate bandwidth is a really stunning figure,
being able to forward packets between 25 Ethernets at full tilt.

If it really can route 22MB/s then whatever processor card is inside a CISCO
router must have a bus that is at least four times as fast as an ISA bus,
nearly as fast as an EISA bus, and it must have a diabolically fast memory
subsystem.

22MB/s cannot be achieved by most :-) 386 memory subsystems for pure string
copies.  Try copying a 1KB string from one buffer to another for a few
thousand times and then calculate the memory bandwidth.

Just to give people an idea of the numbers involved, here is a transcript of
a quick test done on the local DECsystem 5830, which has got a screamingly
fast (20-25MIPS) 25Mhz R3000 CPU:

    Script started on Mon Sep  2 16:04:31 1991
    $ cat copy.c
    #define N (1024/sizeof (int))
    
    int from[N],to[N];
    
    main()
    {
	register long	k = 64*1024;
    
	while (k--)
	    bcopy(from,to,sizeof to);
    }
	
    $ cc -o ./copy -O copy.c
    $ time ./copy
	    8.3 real         7.0 user         0.1 sys  

I read 7 seconds for copying 64MB, or almost 10MB/s. This is reading and
writing, with no processing. Maybe this 25Mhz R3000 CPU can beat a CISCO
router bandwidth only if one does simply reads from its own cache, so let's
have a look at just reading:

    $ cat read.c
    #define N (1024/sizeof (int))
    
    int frame[N];
    
    main()
    {
	register int	i;
	register volatile	v;
	register long	k = 64*1024;
    
	while (k--)
	    for (i = 0; i < N; i++)
		v = frame[i];
    
	return v;
    }
	
    $ cc -o ./read -O read.c
    $ time ./read
	    4.8 real         4.1 user         0.0 sys  
    $ ^]
    script done on Mon Sep  2 16:05:55 1991

That's a tad over 4 seconds for 64MB, or 16MB/s.

I'd be truly amazed if the CISCO router could handle 22.5MB aggregate
bandwidth, even if just for forwarding, unless it has inside a processor
memory combination substantially faster than a 5830, or it has a wormhole
chip. Somehow I think both hypothesises are somewhat improbable.

> I suspect that the quoted Cisco rates are for minimum (50
> byte) or fairly small packets, netting somewhere in the neighborhood of
> 100KBytes/sec. 

I suspect that too, so that at 15k packets/s we'd get about 0.8MB/s for
forwarding. Looks more like it. CISCO may not have gotten a Shannon
dispensation after all.

The original poster should tell us, as I said, what kind of packet size and
frequency distribution he expects and so on.
-- 
Piercarlo Grandi                   | ARPA: pcg%uk.ac.aber@nsfnet-relay.ac.uk
Dept of CS, UCW Aberystwyth        | UUCP: ...!mcsun!ukc!aber-cs!pcg
Penglais, Aberystwyth SY23 3BZ, UK | INET: pcg@aber.ac.uk