[c-nsp] availability

[Vincent C Jones]

On Sat, 2010-02-20 at 17:16 +0200, Saku Ytti wrote:
> On (2010-02-19 11:44 -0500), My Name wrote:
> 
> > Does anyone have information concerning calculating
> > network availability based on a network design?
> > 
> > For example, is redundant P and PE routers more available statistically
> > than single P and PEs with redundant route processors, etc .....?
> > 
> > I am looking to input network design parameters and produce an
> > availability/probability number?
> > is there such an animal?
> 
> If not anything else, this might point to some directions or help with more
> precise questions: http://iplu.vtt.fi/
> 
> Unfortunately some of the material is finncryp() and it is also very high
> level, done by professional academics on goverment grands, some co-workers
> of mine have met these people to give them some real life data to work with
> and they reported that it was really hard to follow what they were doing,
> so expect formulae etc.
> I think in real life engineers know this stuff intuitively.
> 
> I sometimes wonder if high network quality even pays off, at least our
> claimed SLA compensations are very low so it would be hard to justify any
> CAPEX increase to increase quality. Rather it would seem that network
> quality can be decreased if it means we can be more competitive or have
> higher margins.
> Spending money on brand, creating high perceived quality might be wiser
> than actually trying to increase quality, since actual quality is quite
> hard to measure.
> But of course it is much more fun and satisfying to create the best network
> you can. Too bad majority of customers claim they want quality, but seen to
> choose cheapest option from market, perhaps even the worst product is good
> enough.

There is an overview of the math involved in calculating the
availability of a network design in the first chapter of my book "High
Availability Networking with Cisco." While quite simple in theory, it
can be notoriously inaccurate in the real world due to GIGO (if the
component availabilities you start with are garbage, so will the
calculated overall availability). The book is out of print, but is
frequently available on Amazon at bargain prices. Feel free to contact
me off-list if you have trouble finding a copy.

Do keep in mind Saku's warnings. Over the years I've seen a number of
networks where the added redundancy decreased availability. If not
implemented properly, the only thing which will be improved are the
vendor's profit.

Good luck and have fun!
-- 
Vincent C. Jones
Networking Unlimited, Inc.
Phone: +1 201 568-7810
V.Jones at NetworkingUnlimited.com