[ofa-general] Re: [PATCH] opensm/osm_node_info_rcv.c: create physp for the newly discovered port of the known node

[Yevgeny Kliteynik]

Hi Sasha,

Sasha Khapyorsky wrote:
> Hi Yevgeny,
> 
> On 23:31 Wed 18 Feb     , Yevgeny Kliteynik wrote:
> 
> [snip...]
>> Good point.
>> I'll repost the patch when we finish discussing it.
> 
> Let's go this way now. Please resend the patch.

Will do.

> After looking closer into scenario with SwithInfo/PortInfo race I'm
> thinking about two optimizations there:
> 
> 1. Initialize all switch ports (and not only local and port 0) right on
> first NodeInfo receiving (via osm_node_new()) - this makes your patch
> unnecessary, but it is a bigger change which will definitely require some
> heavy testing, so it is fine IMO to do it subsequently.
>
> 2. Request PortInfo for all switch ports right on first NodeInfo
> receiving (not wait for SwitchInfo), just in parallel with SwitchInfo
> request. This should simplify subnet discovery flow and speed it up.
> And also this will require some heavy testing...
> 
> What do you think about (1) and (2). Could you see any disadvantages?

I don't see any.

The first option looks shorter and somewhat more safe, but the second
option might speed up the discovery a little bit. I'm OK with both options.
In any case, this will have to be seriously tested.

-- Yevgeny

> Sasha
>