Hello,
Just was trying out 82599 VFs per the
subject and I'm finding that traffic
for multicasts subscribed on any VF
of a physical interface is delivered
to every VF.
Seeing this with 'tcpdump -p' (not
promiscuous). So even though the
kernel will discard the duplicate
traffic when delivering to UDP
sockets, the extra load on the
kernel defeats the purpose I have
in mind.
Is this duplication always present,
or is it a function of the older
kernel in use? Do any knobs or
dials I may have missed address
the problem (did read README
and search the issue).
Note VF are working fine, aside
from the UDP multicast packet
duplication. The multicast
MAC subscriptions are visible
in /proc/net/dev_mcast and
are nicely distributed to
different VFs.
Thanks for any help or advice.
Tech details below.
David
older server, older 82599
SuperMicro 1026T-URF w/ X8DTU mainboard
5520 northbridge
ICH10R southbridge
dual X5560 cpu
BIOS 2.1c 8/3/12 (last)
X520-DA2 dual-82599ES SFP+ 10G
CentOS 5.11
kernel 2.6.18-398.el5
ixgbe 3.22.3
ixgbevf 2.15.3
boot: intel_iommu=on iommu=pt
modprobe ixgbe max_vfs=16,16 RSS=4,4
-----
P.S. Experienced a minor quirk in
configuration. Had to set the
VF MAC addresses twice, once
before 'ifup' and once after
'ifup' or the VFs would jam
with "Spoofed packets detected"
errors.
------------------------------------------------------------------------------
Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
from Actuate! Instantly Supercharge Your Business Reports and Dashboards
with Interactivity, Sharing, Native Excel Exports, App Integration & more
Get technology previously reserved for billion-dollar corporations, FREE
http://pubads.g.doubleclick.net/gampad/clk?id=157005751&iu=/4140/ostg.clktrk
_______________________________________________
E1000-devel mailing list
E1000-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/e1000-devel
To learn more about Intel® Ethernet, visit
http://communities.intel.com/community/wired
