Speeds and feeds

Andrew J. Luca andrewluca@mediaone.net
Sun, 7 Jun 1998 12:01:07 -0400


I want to point out that the issues that I presentedare based in reality as
implemented by companies who can not (read lose millions of dollars per hour
or minute) tolerate down time - ever.  We ran multi-user UNIX systems like
they were mainframes with target annual downtimes approaching zero (99.97%).
We achieved those numbers for six years running - the entire lifetime of my
organization at that company.

	You may not agree with running dynamic routing protocols on a firewall -
and in most cases I do not either - however, firewalls will not scale to
support electronic business in their current incarnation.  What is missing
today is an additional layer of fault tolerance and redundancy.  The first
of these layers comes from a software (or possibly hardware) high
availability package.  This brings a large gain of availability to the
picture.  However, without some type of dynamic fail-over for the network
links (again, I point out that this is your weakest link) you still can't
guarantee the highest levels of availability.  I would assert that with
proper controls implemented, it is possible to run a dynamic protocol on a
firewall with little to no additional risk.  The problem with dynamic
routing protocols come from using simple protocols (like RIP) without proper
controls.  Running a more configurable protocol (like OSPF) with controls in
from of t firewall (like filtering routers) and control over the routing
package (like a stripped gated) provides this.

	Your solutions for multiplexing DS-1's will scale to any size that you can
reasonably imagine.  After all, DS-3's are simply lots of DS-1's multiplexed
together!  Your only limit is the number of interfaces that your router has.
Since this has taken such a theoretical bent, I would argue that this could
be any number.  (Not a practical number, though ;-))  Looking into a capped
(or channelized or whatever) type of DS-3 can be a good choice for a
company.  The time that you spend installing 2 Mbps worth of bandwidth today
just to upgrade to 4 or 6 nine months from now can be more expensive than
the additional costs.  Many providers in metro areas today are pushing
clients to consider channelized DS-3 capacity as a way to justify local loop
upgrades.  A DS-3 implementation that I looked at four months ago for two
DS-1's vs. a 4Mbps DS-3 put the DS-3 solution far ahead of the DS-1's.  You
need to consider what you could potentially do with the remaining bandwidth
of the DS-3 before making a decision like that.  You can let the telco
channelize the DS-3 for you and only take the segment that you need.  I bet
that they have a use for the extra bandwidth even if you don't.  This, of
course, would not apply if you lived in Wyoming.  But, then I already sataed
that we were talking about metrareas.

	Also, from experience, the big providers are the only way to ensure that a
large company is always connected to the Internet.  The mom & pop shops are
drying up these days.  The big guys are the only ones that have enough
survivability to ensure that they can meet demanding SLA's.  If you think
that any type of significant electronic commerce will take off without this,
you are dead wrong.  There are few if any large scale implementations of
true electronic commerce today.  (The idea that electronic commerce is
sending your credit card over the Internet is silly - we never argued that
calling L.L. Bean with our credit card was telephony commerce!)  One of the
biggest fears in corporate America is whether electronic commerce can
actually support a business - that means in every aspect of the model not
just in one area.

-Drew