Thinking Methodically about building a PoC

Mon Jun 13 15:40:25 UTC 2016

Agreed, the margin of growth never seems to stay consistent or be what you predict.

Instead of asking "What do you want?" (or taking their specs).  I have lots of ideas of what I want but maybe not what I need.

Ask these questions, they've gotten me better answers and have allowed me to do the hardware/software framework not the end user/client.

"What are you trying to sell?"

"What are you trying to do?"

Brian

________________________________
From: NANOG <nanog-bounces at nanog.org> on behalf of Matthew Petach <mpetach at netflight.com>
Sent: Sunday, June 12, 2016 11:20 PM
To: Roland Dobbins
Cc: NANOG list
Subject: Re: Thinking Methodically about building a PoC

On Sun, Jun 12, 2016 at 9:49 PM, Roland Dobbins <rdobbins at arbor.net> wrote:
>
> On 13 Jun 2016, at 8:52, Kasper Adel wrote:
>
>> 2) Do some planning and research first.
>
> This.
>
> -----------------------------------
> Roland Dobbins <rdobbins at arbor.net>
>

We never design in a vacuum.  There's always some
target we're  designing towards.  Testing is no different.
Think about what it is you'll need to support.  Look at
historical numbers related to those features/capabilities.
Yes, as the stork market keeps reminding us, past
performance is no guarantee of future results...but at
the same time, those who don't learn from the past
are doomed to re-implement it...poorly.

So, when we test, we look at protocols we've already
been running for years, and then we look at the growth
curves we've seen in those protocols over the past X years,
where X is approximately the estimated lifespan of the
hardware in question.  So, if the current router platform
you're looking to replace has been in place in your
network for 8 years, and you're testing the next
generation for BGP route scaling, look at what
the global BGP table size was 8 years ago,
and look at where it is today; work out the percentage
growth curve for it; then take the current BGP table
size, apply the same compound growth percentage
to it for the next 8 years, and you'll come up with a
reasonable idea of the scale you'll need the box
to handle over its lifetime.  Test that; then, to give
yourself a margin of error, double the number, and
test again.  That way you have a realistic idea of
whether it can support your current growth rate,
and whether it can support your growth if the
growth rate is 1.4x what you expect.

Do those calculations for each of the protocols
under test, and you'll be able to come up with
a reasonable testing profile that's supportable
based on historical information, rather than flights
of fancy.

Hope that helps!

Matt