[ofa-general] Re: [PATCH 3/7 V2] osm: QoS policy C & H files

Sasha Khapyorsky sashak at voltaire.com
Mon Aug 27 18:12:54 PDT 2007


On 20:37 Mon 27 Aug     , Hal Rosenstock wrote:
> On 8/27/07, Sasha Khapyorsky <sashak at voltaire.com> wrote:
> > On 12:35 Mon 27 Aug     , Hal Rosenstock wrote:
> > > >
> > > > Thanks.
> > > >
> > > > > >>>> +        cl_list_t group_list;   /* list of group names (strings) */
> > > > > >>>> +        cl_list_t across_list;  /* list of 'across' group names (strings) */
> > > > > >>>> +        cl_list_t vlarb_high_list;      /* list of num pairs (n:m,...), 32-bit
> > > > > >>>> values */
> > > > > >>>> +        cl_list_t vlarb_low_list;       /* list of num pairs (n:m,...), 32-bit
> > > > > >>>> values */
> > > > > >>> Why cl_list for VLArb? it should be short fixed length arrays?
> > > > > >>  Right.
> > > > > >>  Since the actual VLArb setup is not implemented yet, I didn't see
> > > > > >>  this obvious thing.
> > > > > > But it should be implemented. Right?
> > > > >
> > > > >  Sure, but not for OFED 1.3 - we have a feature freeze in 11 days.
> > > >
> > > > Then we will have two QoS managers in parallel, I don't like this too
> > > > much.
> > >
> > > Isn't this needed ? What about LASH ? How is that supported ?
> >
> > Do you mean potential conflict between LASH and QoS in terms of SL/VLs?
> 
> Yes, that and setting VLarb tables.

Good point. It is needed to provide at least some level of QoS/LASH
co-existence.

> > > I
> > > thought there were extra options or the like to enable the higher
> > > level QoS functions (manager) ?
> >
> > There is already '--qos' option, do you think it is not enough?
> 
> Will it continue to mean what it does at OFED 1.2 ?

Would be nice to not change if not really necessary.

> How is the high
> level QoS distinguished from the existing lower level QoS in terms of
> starting OpenSM ?

Do you think it could be useful as separate option? I'm not sure yet,
but probably yes.

Sasha



More information about the general mailing list