I am going to run BGP on 2 x 3750 stacked with 2 diff PEs from a Single upstream. The physical connectivity will be as below
switch a - stacked - switch b
PE A PE B
It will be an active - passive setup. Local preference and MED will be used to control incoming and outgoing traffic respectively on the switches. On the PEs end, I suspect my provider will using local preference and AS-prepend to control the active and passive setup.
Both of the link will be 100Mb and my provider will be advertising just a default route to the switches. I wonder will there be any performance issue running this type of setup on the 3750? The IOS on the 3750 is capable of supporting BGP.
I do not have personal experiences with the setup you are about to deploy. Nevertheless, I believe that there will be no significant performance issues if this solution is deployed properly.
My reasoning is that the only major causes of a possible performance issue are the BGP process itself and the size of the routing table. If the BGP table you receive from your PEs is reasonably small and stable then the BGP process itself running on the 3750 will be fairly quiet, consuming only a fraction of system resources (CPU and RAM). I would personally suggest aggresively filtering the routes received from the PE routers so that only selected networks of interest and/or the default route is accepted from PEs - you certainly do not want to have the full BGP table uploaded to you. Also take into consideration that the routes in the routing table are compiled and stored into the TCAM memory of your 3750 whose space is quite precious. If some of the routes do not make it into the TCAM then packets to these networks will be punted to the CPU for process routing, possibly overwhelming it.
So, I believe it can be done nicely without significant performance issues if the BGP is adequately controlled.