[nycbug-talk] Can RTSP and dumb switches coexist?

[Jesse Callaway]

On 10/18/07, Alex Pilosov <alex at pilosoft.com> wrote:
> On Thu, 18 Oct 2007, Jonathan Stewart wrote:
>
> > As the subject asks based on what I have read so far (including chunks
> > off the IEEE spec) it appears that RTSP can coexist with dumb switches
> > as long as the dumb switches themselves don't form loops.  Can someone
> > confirm or refute this for me please.
> a) I suppose you mean RSTP (rapid spanning tree protocol). RTSP is voip
> blah.
>
> b) dumb switches *must* comply with 802.1d (the STP protocol) anyway.
> RSTP is backward compatible (theoretically), and it should (theoretically)
> work fine, with loops or without.
>
> c) the above being said, I encourage my competitors to rely on any
> incarnation of STP to detect loops.
>
> d) if above is too dense, explanation:
> 1) You *really* do not want loops in your l2 network. doubly so if you are
> mixing STP protocols.
>
> 2) Unless you understand *exactly* how all variations of STP work, and how
> they interact with each other, you *will* shoot yourself in the foot.
>
> 3) When STP breaks, it breaks in a very strange, and unstable ways, and
> takes down your network completely.  Troubleshooting it, when your network
> is down, is not for weak-minded.
>
> 4) Just don't do it (tm). This is what happens when STP breaks:
> http://www.snwonline.com/storage_knowledge_center/all_systems_down_03-03-03.asp
>
> 5) Use layer 3 if you need redundancy (and loops).
>
> The above being said, we (pilosoft) use STP/RSTP/PVST+/etc...But, the
> thought about touching my configs makes hair raise on top of my head.
> It'll be shortly ripped out and replaced with proper l3 stuff.
>
> -alex
>


Cool story. Little did I think it would be a life and death disaster.
I've screwed up before, but to take down 1/4 of the boston health
network is mind blowingly horrible. It certainly is a hairy fuzzy line
between layers 2 and 3 sometimes. One that always needs reevaluation
in any buildout worth a damn.
Of course it really depends on the exact needs of the network, but
what sort of a replacement solution would you implement? Layer 3 aware
switches using VRRP or the like? I'm assuming the original problem is
link-layer redundancy, and not access to alternate routes.
This is totally the wrong list, and I am just dragging this further
into obscurity.

-jesse