[ofa-general] RE: [PATCH] [10 of 10] [REVISED] mesh analysis - integrate into lash core
Robert Pearson
rpearson at systemfabricworks.com
Mon Dec 15 12:08:31 PST 2008
The p3 fix impacted this one as well. The rest are unchanged.
-----Original Message-----
From: Robert Pearson [mailto:rpearson at systemfabricworks.com]
Sent: Wednesday, December 10, 2008 12:56 PM
To: 'Sasha Khapyorsky'; 'general at lists.openfabrics.org'
Subject: [PATCH] [10 of 10] [REVISED] mesh analysis - integrate into lash
core
Sasha,
Here is a revised mesh patch #10 that incorporates changes based on your
comments.
This patch
- hooks mesh code into lash
- replaces sw->phys_connections by the equivalent switch->node->links
- replaces sw->num_connections by the equivalent
switch->node->num_links
- replaces sw->virtual_physical_port_table by
switch->node->links[]->ports
When the do_mesh_analysis flag is not set there is no change to the function
except To replace the variables with variables in node that have the same
size. In this Case the port table in link_t will always have just one port.
When the do_mesh_analysis flag is set multiple physical links will collapse
to a Single logical link with a port list with more than one element.
- rewrote connect switches to use variables in node
- in log Lane requirements (%d) exceed available lanes (%d)
Arguments were reversed, fixed
- compute physical egress port in routine get_next_port
Which will use round robin if there are more than one
Physical links between switches
Regards,
Bob Pearson
-------------- next part --------------
A non-text attachment was scrubbed...
Name: p10
Type: application/octet-stream
Size: 11387 bytes
Desc: not available
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20081215/a00d00f3/attachment.obj>
More information about the general
mailing list