[c-nsp] Data Center switch replacement

[Brian Spade]

Hi,

On Thu, Dec 17, 2009 at 6:42 PM, Randy McAnally <rsm at fast-serv.com> wrote:

>
> > How about you individually move each connection over to the secondary
> > switch one at a time.  This should only be a 30 second downtime
> > window per port, I'd think?  Once you've migrated everybody off of
> > the primary switch, pull it, upgrade it and then move everybody back
> > one-by-one?  This would minimize everybody's downtime and I think
> > would go over better with your clients.  Plus, you can drag out the
> > upgrade over time rather than an "all or none" scenario.
>
> Agreed.  What if something goes wrong or takes longer than expected --
> wouldn't you like to know by the time you've moved the first cable and not
> after the original switch is completely offline and de-racked?
>

+2

For example, install new switch, make it's connections to the existing AGG
layer but also interconnect it to your existing ACC layer, pre-configure
port assignments from current access switch 2 to new switch 2, move
connections one at a time off of your current access switch 2 to new switch
2.   Remove old access switch 2, install new access switch 1,
rinse-and-repeat.

Single-homed devices will have down-time from however long it takes you to
physically move the port plus your CAM timeout.  If you can do this fast
enough you might not reset some TCP connections.

However, if there is a lack of infrastructure for this... fix the servers or
determine how long the hosts will be down :-)

/bs