Roland Dreier wrote:
I take this back. These stats are a subset of the IP and TCP MIBs.
I'd like to get something out in 2.6.27/ofed-1.4 for support reasons without
opening the SNMP can of worms. So I'll define a standard struct
iw_protocol_counters and include the chelsio stats. Other providers can
expand if they care. I'll also add the core stuff to have sysfs.c create
the attribute files, one per stat and query the device to get the values via
a device get_iw_stats() call like you propose. The stats can live in
/sys/class/infiniband/$device/protocol_counters/iwarp/*.
How does that sound?
Sounds great except I'm not thrilled about the exact sysfs pathname...
I guess iWARP devices always have one port (??) (physically multiport
NICs show up as multiple devices? don't have good access to my system
with a two-port rnic to check) The only point of all that being that
$device/ports/$port may not make total sense.
The TCP and IP protocol stats are global to the device. Just like the
native stack's TCP/IP stats aren't per-nic, we can't report offload TCP
stats per-port. Yes? Note per-port stats are reported via ethtool -S
for iwarp devices (L2 stats).
Steve.
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
