On Fri, Dec 21, 2001 at 01:34:44PM -0700, Irwin Lazar reportedly typed:
>What, if anything, makes a multi-vendor (wide-area) network
>successful and worth the risks over the "safe" single-vendor
>network nobody gets fired for buying (you can probably guess
>what vendor Powers my network now).
I like thinking of where an organization wants to be on a risk/reward
spectrum. Newer and/or point solution vendors exist to leapfrog the status
quo and give you an advantage, which they do very well. This is
particularly true on the optical side, where lasers, components, and
software have all changed dramatically. The price you pay is risk. If you
don't have capacity problems or take an incumbent role, then you would
prefer a defensive, conservative strategy favoring the single-vendor
solution.
You have to quantify those comments a little bit more.
In an IP network, when you mix vendors for a particular application, you
do not gain speed at all, but in fact, quite the opposite.
Allow me to provide an extreme example.
Let's say you select vendor A for it's MPLS VPN capabilities on the edge.
Then you decide that you like the density for a certain customer size
in vendor B better. You implement both vendors for this application, but in
doing so, you must now use only the variety of MPLS-based VPNs that are
compatible between vendors. Any particular VPN variety that provides unique
capabilities will not be available to you until both vendors implement a
compatible, possibly even standards-based version.
This liability extends into your core as well. A mixed-vendor core cannot
implement several advanced MPLS pre-standard capabilities, such as
fast-reroute, even though interoperability issues have been worked out for
most basic MPLS functions.
I use MPLS as an example, but any vendor specific pre-standard capability
would apply, such as UTI, SRP(DPT), cdp, etc.
Your functionality becomes least common demoninator. Your product capability
is limited by the slowest of the two vendors to implement, not the fastest.
You can mitigate this problem by carving up your network into roles. Core,
Internet edge, VPN edge, etc. Then you can pick the best of breed within
that role, but all roles need not be a single vendor. Obviously,
substantial testing is necessary...YMMV.
Dave