Colored Backgrounds

[Steve Lamb]

Monday, December 13, 1999, 2:42:56 PM, David wrote:
> Thanks for the explanation but I'm still not convinced that there are any
> physical limits to capacity which cannot be overcome with appropriate
> funding.

    There are.  This is an extreme Libertarian viewpoint.  While I subscribe
to a lot of their philosophy I won't ever concede that an ininite amount of
cash will solve all problems.

> In a previous posting you mentioned that your current employer has yet to
> turn a profit. Unless demand and capacity keep on expanding he never will
> and if he doesn't ... well, who is going to run the PMmail mailing list ...?

    That is because we're also not utilizing all of the possible sources of
revenue that we can.

> If there ain't enough processing capacity buy some more.

    And hook them together how?

> If there ain't enough storage capacity buy some more discs

    And hook them together how?

> If the pipe ain't big enough get some more fibre optic - it's dirt cheap and
> getting cheaper.

    And hook them together how?

> Or am I being hopelessly naive ...

    You're being hopelessly naive.  I don't mean that in a rude way, but trust
me, it isn't as simple as just throwing more money at the problem.  I singled
out the above three to show something.

    Here are two computers on a network.

C1-----C2

    Now 3.


C1------C1
 \      /
  \    /
   \  /
    \/
    C3

    Now 4.  Wait, we need to add a device to that

C1    C2    C3   C4
 |     |     |    |
 |    ---------   |
 +----[  Hub  ]---+
      ---------

    Now, note that a hub is just a dumb device.  Anything that is sent to it
on one port is sent to all the other ports.  This places a limit on the number
of ports you can place on the hub since the total bandwidth (let's take
10Mbit) is now shared between 4 devices.  Put 400 on there and performance
degrades to being unusable from collisions.

    So now we have, say, 4 hubs with 10 machines each.  40 machines.  We need
a switch to handle the traffic.

H1    H2    H3   H4
 |     |     |    |
 |    ---------   |
 +----[ Switch]---+
      ---------

    A switch knows, basically, which mac addresses are on which segments and
routes accordingly.  It stops information from one segment spilling over to
the other, reducing traffic.  However, it can only do it so fast.  But it
doesn't route, so we need a router.

H1    H2    H3   H4
 |     |     |    |
 |    ---------   |
 +----[ Switch]---+
      ---------
          |
      ---------
      [ Router]
      ---------

    A router knows where to send packets.  Ah, but here's the catch, it needs
a route table to know where to send the packets to.  That takes processor time
and memory.

    Now, scale that simple network from 400 people to 4,000.  4,000,000.
4,000,000,000.  It doesn't work.  Why?

    How do the routers know where to route all of the packets to?  The tables
are too large.

    This is, of course, a basic illustration of one problem.  You can't just
throw more computers at the problem because they need to connect to one
another.  You can only have so much bandwidth over each segment.  Each switch,
router, whatever, only has so much processing power to do its job.  Add
another one into the mix and you're facing problems of traffic on each segment
again.  It is hard to express that the limits are there but they are there.

-- 
         Steve C. Lamb         | I'm your priest, I'm your shrink, I'm your
         ICQ: 5107343          | main connection to the switchboard of souls.
-------------------------------+---------------------------------------------