[Tfug] Ethernet frame "immutables" wrt switch silicon

[Bexley Hall]

Hi Zack,

On 5/15/2013 12:44 PM, Zack Williams wrote:
> On Tue, May 14, 2013 at 11:24 PM, Bexley Hall<bexley401 at yahoo.com>  wrote:
>
>> If you disassemble a commercial switch, you'll see that the switch
>> is implemented (typ) in several "identical" smaller bits of silicon.
>> E.g., a 12 port switch may be three 4-port "chips" internally.
>
> Depends on the topology and technology.  I popped the top of my 24-port
> 20Gbit/s infiniband switch when trying to determine which of the 2 power
> supplies had died on it, and it had all the ports into one enormous chip.

Chances are, they use a crossbar topology inside the switch.
As speeds go up, the cost of "sharing" (a bus, memory, etc.)
becomes too expensive.  Easier to throw more silicon at it
(but that means higher level of integration).

> But in general, multiple chips with a high speed interconnect is the more
> conventional method.

It will be interesting to see where the "discrete" vs. "integrated"
approaches diverge wrt network link speeds.  Going "off chip"
always has some large speed/power penalties so we actually may
be at that point with Gb and 10G speeds!  In which case, we'll
start seeing smaller -- fewer ports -- switches as the cost
cutting measure (along with more "unmanaged" features).

[To think, when I started out in this industry, 125MHz was as
close to "light" as we thought practical!  :-/  And the power
supplies used were several times larger than modern PC's!!
What a cool vocation/avocation.  Imagine how boring it must be
to work in an industry where things slowly/*never* change...  :< ]