I would appreciate it, if you could take this serious.The topic is related to RSS. If you do not know what's it, please consult next level, please.
what the manual says.....
The x710 supports multiple queues to spread the incoming traffic accross the cpus, per magic flowdirector.
It should be set up under linux running the command ethtool like
ethtool -L eth0 combined 7
should use 7 rx/rx combined queues and distribute the traffic up to 7 irqs.
if 7 irqs are smp_affinitiy'iated to 7 cores, we can say, each queue will be handled by one core.
That was a short summary of intels manual, intel's saying how the adapter should work.
again, what's happen in reality...
in our real scenario we have 1 cpu with 8 cores.
if we run "ethtool -i ethX combined 8" each queue will end up to one core.
if we run "ethtool -i ethX combined 7" there will be 8 queues, balanced to one irq/core.
Same happend if we run this with 2 or 3 or 4 ...
We can conclude:
If the count of queues is not equal to the count of cores, it shows the same misbehavior.
It just create one queue.
So this is not normal. this is not whats written in the manual.
Please tell me what's wrong. The manual or the driver or the card?
02) Intel(R) Ethernet Controller X710 6.01(6.01) 1572 00:004 Update not available
for 10GbE SFP+
04:00.0 Ethernet controller: Intel Corporation Ethernet Controller X710 for 10GbE SFP+ (rev 02)
04:00.1 Ethernet controller: Intel Corporation Ethernet Controller X710 for 10GbE SFP+ (rev 02)
04:00.2 Ethernet controller: Intel Corporation Ethernet Controller X710 for 10GbE SFP+ (rev 02)
04:00.3 Ethernet controller: Intel Corporation Ethernet Controller X710 for 10GbE SFP+ (rev 02)
description: Ethernet interface
product: Intel Corporation
vendor: Intel Corporation
physical id: 0.1
bus info: pci@0000:04:00.1
logical name: ethX
width: 64 bits
capabilities: pm msi msix pciexpress vpd bus_master cap_list rom ethernet physical fibre
configuration: autonegotiation=off broadcast=yes driver=i40e driverversion=2.7.26 duplex=full firmware=6.01 0x800036e4 1.1861.0 ip=a.b.c.d latency=0 link=yes multicast=yes speed=10Gbit/s
resources: irq:16 memory:91000000-917fffff memory:92c08000-92c0ffff memory:fb200000-fb27ffff memory:90600000-907fffff memory:92ca0000-92d1ffff
ip, serial and logical name were modified.
firmware-version: 6.01 0x800036e4 1.1861.0
... but look. i can not see that the effect was system or kernel dependend. the chip/card is capable of balancing to the number for cores of the cpu.
ethtool just tells the driver the amount of queues. it looks like a bug in the latest driver.
we use quad port sfp+ card AOC-STG-I4S
we have asked them before.
they recommended broadcome series instead. AOC-STG-b4S
[~]# more /etc/redhat-release
CentOS release 6.5 (Final)
[~]# uname -a
Linux dummy 3.13.2-4.el6.x86_64 #1 SMP Wed Feb 26 13:31:21 CET 2014 x86_64 x86_64 x86_64 GNU/Linux
[~]# ethtool -k eth0
Features for eth0:
tx-checksum-ip-generic: off [fixed]
tx-checksum-fcoe-crc: off [fixed]
tx-scatter-gather-fraglist: off [fixed]
udp-fragmentation-offload: off [fixed]
large-receive-offload: off [fixed]
rx-vlan-filter: on [fixed]
vlan-challenged: off [fixed]
tx-lockless: off [fixed]
netns-local: off [fixed]
tx-gso-robust: off [fixed]
tx-fcoe-segmentation: off [fixed]
tx-mpls-segmentation: off [fixed]
fcoe-mtu: off [fixed]
loopback: off [fixed]
rx-fcs: off [fixed]
rx-all: off [fixed]
tx-vlan-stag-hw-insert: off [fixed]
rx-vlan-stag-hw-parse: off [fixed]
rx-vlan-stag-filter: off [fixed]
l2-fwd-offload: off [fixed]