[ofa-general] Re: [OpenSM] updn routing performance fix???

[Sasha Khapyorsky]

On 01:46 Sat 01 Mar     , Sasha Khapyorsky wrote:
> On 09:58 Fri 29 Feb     , Albert Chu wrote:
> > 
> > What you're trying to do is calculate "ignore_existing_lfts" when the port
> > trap is received rather than during routing later on?
> 
> Not trap but when PortInfo is received during subnet discovery phase of
> sweep (before routing configuration).

I started to test this I saw that it solves the problem only partially.

For example in such topology:

         sw3
       /     \
 - sw1         sw2 ->
       \     /
         sw4

Let's disconnect link between switches 2 and 3, paths to arrow direction
at switch 1 will be rerouted via sw4. Now let's restore the link -
switches 2 and 3 will be enforced to rebalance routing paths, but sw1
will not and all traffic will go via sw4.

Looks we need to set ignore_existing_lfts globally when any switch was
reconnected in a fabric.

Sasha