[ofa-general] [PATCH] Opensm: switchbalance console option

[Al Chu]

Hey Sasha,

On Thu, 2008-03-13 at 20:47 +0000, Sasha Khapyorsky wrote:
> On 15:37 Wed 12 Mar     , Al Chu wrote:
> > Hey Sasha,
> > 
> > Forgot to run this through osm_indent.  Here's the cleaned up patch.
> > 
> > Al
> > 
> > On Wed, 2008-03-12 at 09:57 -0700, Al Chu wrote:
> > > Hey Sasha,
> > > 
> > > Here's a patch that does the previously discussed switchbalance console
> > > option.  Algorithmically it does pretty much the exact same thing as the
> > > check_lft_balance script, but everything is faster of course b/c opensm
> > > already knows everything.
> > > 
> > > Al
> > > 
> > > _______________________________________________
> > > general mailing list
> > > general at lists.openfabrics.org
> > > http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
> > > 
> > > To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
> > -- 
> > Albert Chu
> > chu11 at llnl.gov
> > 925-422-5311
> > Computer Scientist
> > High Performance Systems Division
> > Lawrence Livermore National Laboratory
> 
> > From ffdba2635aeea1d1be58c8966f8f6137bb048dea Mon Sep 17 00:00:00 2001
> > From: Albert L. Chu <chu11 at llnl.gov>
> > Date: Wed, 12 Mar 2008 15:34:49 -0700
> > Subject: [PATCH] add switchbalance command to console
> > 
> > 
> > Signed-off-by: Albert L. Chu <chu11 at llnl.gov>
> 
> Applied. Thanks. Small question is below.
> 
> [snip...]
> >  
> > +static void switchbalance_check(osm_opensm_t * p_osm,
> > +				osm_switch_t * p_sw, FILE * out, int verbose)
> > +{
> > +	uint8_t port_num;
> > +	uint8_t num_ports;
> > +	const cl_qmap_t *p_port_tbl;
> > +	osm_port_t *p_port;
> > +	osm_physp_t *p_physp;
> > +	osm_physp_t *p_rem_physp;
> > +	osm_node_t *p_rem_node;
> > +	uint32_t count[255];	/* max ports is a uint8_t */
> > +	uint8_t output_ports[255];
> > +	uint8_t output_ports_count = 0;
> > +	uint32_t min_count = 0xFFFFFFFF;
> > +	uint32_t max_count = 0;
> > +	unsigned int i;
> > +
> > +	memset(count, '\0', sizeof(uint32_t) * 255);
> > +
> > +	/* Count port usage */
> > +	p_port_tbl = &p_osm->subn.port_guid_tbl;
> > +	for (p_port = (osm_port_t *) cl_qmap_head(p_port_tbl);
> > +	     p_port != (osm_port_t *) cl_qmap_end(p_port_tbl);
> > +	     p_port = (osm_port_t *) cl_qmap_next(&p_port->map_item)) {
> > +		uint16_t min_lid_ho;
> > +		uint16_t max_lid_ho;
> > +		uint16_t lid_ho;
> > +
> > +		/* Don't count switches in port usage */
> > +		if (osm_node_get_type(p_port->p_node) == IB_NODE_TYPE_SWITCH)
> > +			continue;
> > +
> > +		osm_port_get_lid_range_ho(p_port, &min_lid_ho, &max_lid_ho);
> > +
> > +		if (min_lid_ho == 0 || max_lid_ho == 0)
> > +			continue;
> > +
> > +		for (lid_ho = min_lid_ho; lid_ho <= max_lid_ho; lid_ho++) {
> > +			port_num = osm_fwd_tbl_get(&(p_sw->fwd_tbl), lid_ho);
> > +			if (port_num == OSM_NO_PATH)
> > +				continue;
> > +
> > +			count[port_num]++;
> > +		}
> > +	}
> > +
> > +	num_ports = p_sw->num_ports;
> > +	for (port_num = 1; port_num < num_ports; port_num++) {
> > +		p_physp = osm_node_get_physp_ptr(p_sw->p_node, port_num);
> > +
> > +		/* if port is down/unhealthy, don't consider it in 
> > +		 * min/max calculations
> > +		 */
> > +		if (!p_physp || !osm_physp_is_healthy(p_physp)
> > +		    || !osm_physp_get_remote(p_physp))
> > +			continue;
> > +
> > +		p_rem_physp = osm_physp_get_remote(p_physp);
> > +		p_rem_node = osm_physp_get_node_ptr(p_rem_physp);
> > +
> > +		/* If we are directly connected to a CA, its not really 
> > +		 * up for balancing consideration.
> > +		 */
> > +		if (osm_node_get_type(p_rem_node) == IB_NODE_TYPE_CA)
> > +			continue;
> 
> Should this be
> 
> 	if (osm_node_get_type(p_rem_node) != IB_NODE_TYPE_SWITCH)
> 
> ? So routers will be not counted too?

I think you're right.  I should adjust for this in my 'lidbalance'
script and 'check_lft_balance.pl' tool.

I'll post some new patches.

Al

> Sasha
-- 
Albert Chu
chu11 at llnl.gov
925-422-5311
Computer Scientist
High Performance Systems Division
Lawrence Livermore National Laboratory