So I have two VM’s that are slightly different. In one, that actually works.
Both SR-IOV PF on the same physical card (it has two ports). Here’s the
versions:
# working
kernel: 4.6.2-1.el7.custom.x86_64
ethtool version 4.2
$ ethtool -i eth1
driver: ixgbe
version: 4.4.6
firmware-version: 0x80000811, 1.1067.0
expansion-rom-version:
bus-info: 0000:00:05.0
supports-statistics: yes
supports-test: yes
supports-eeprom-access: yes
supports-register-dump: yes
supports-priv-flags: yes
# not working
kernel: 4.6.2-1.el7.elrepo.x86_64
ethtool version 4.5
$ ethtool -i eth1
driver: ixgbe
version: 4.3.15
firmware-version: 0x80000811, 1.1067.0
expansion-rom-version:
bus-info: 0000:00:05.0
supports-statistics: yes
supports-test: yes
supports-eeprom-access: yes
supports-register-dump: yes
supports-priv-flags: no
I’ll try to upgrade the driver version on the one not working, but 4.3.15 is
_fairly_ new, I thought.
—
Derek Ditch
de...@criticalstack.com
GPG: 0x2543A3B5
> On 21Jun 2016, at 14:16, Fujinaka, Todd <todd.fujin...@intel.com> wrote:
>
> That could be a kernel limitation. What kernel are you running?
>
> Todd Fujinaka
> Software Application Engineer
> Networking Division (ND)
> Intel Corporation
> todd.fujin...@intel.com
> (503) 712-4565
>
> -----Original Message-----
> From: Derek Ditch [mailto:de...@criticalstack.com]
> Sent: Tuesday, June 21, 2016 10:57 AM
> To: Skidmore, Donald C <donald.c.skidm...@intel.com>
> Cc: e1000-devel@lists.sourceforge.net
> Subject: Re: [E1000-devel] Support for Symmetric RSS Hashing?
>
> Don,
>
> Thanks for the suggestion. I hadn’t noticed the “hkey” argument to ethtool -X
> before. However, it appears my hardware doesn’t support it (though I could be
> doing it wrong).
>
> sudo ethtool -X ens5 hkey
> 6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A:6D:5A
> Cannot get RX flow hash indir size and key size: Operation not supported
>
> $ sudo lspci -nn -s 05 -v
> 00:05.0 Ethernet controller [0200]: Intel(R) 82599 10 Gigabit Dual Port
> Network Connection [8086:10fb] (rev 01)
> Subsystem: Hewlett-Packard Company HP Ethernet 10Gb 2-port 560SFP+
> Adapter [103c:17d3]
> Physical Slot: 5
> Flags: bus master, fast devsel, latency 0, IRQ 10
> Memory at fea00000 (32-bit, non-prefetchable) [size=1M]
> I/O ports at c080 [size=32]
> Memory at febd0000 (32-bit, non-prefetchable) [size=16K]
> Expansion ROM at feb00000 [disabled] [size=512K]
> Capabilities: [40] Power Management version 3
> Capabilities: [50] MSI: Enable- Count=1/1 Maskable+ 64bit+
> Capabilities: [70] MSI-X: Enable+ Count=64 Masked-
> Capabilities: [a0] Express Endpoint, MSI 00
> Capabilities: [e0] Vital Product Data
> Kernel driver in use: ixgbe
>
> This NIC is passed through as a physical function via SR-IOV, but it doesn’t
> seem to work on the host either.
> —
> Derek Ditch
> de...@criticalstack.com
> GPG: 0x2543A3B5
>
>> On 20Jun 2016, at 11:50, Skidmore, Donald C <donald.c.skidm...@intel.com>
>> wrote:
>>
>> Hey Derek,
>>
>> Why can't you just change the RSS hash key to whatevery you want it to be
>> via ethtool -X?
>>
>> Thanks,
>> -Don Skidmore <donald.c.skidm...@intel.com>
>>
>>
>>> -----Original Message-----
>>> From: Derek Ditch [mailto:de...@criticalstack.com]
>>> Sent: Sunday, June 19, 2016 9:20 PM
>>> To: e1000-devel@lists.sourceforge.net
>>> Subject: [E1000-devel] Support for Symmetric RSS Hashing?
>>>
>>> I’ve been using and recommending Intel NICs for years for network
>>> sensor platforms. I’m a huge user of Bro (http://bro.org
>>> <http://bro.org/>) and also leveraged Suricata for signature-based
>>> detections. Lately, I’ve started working with trying to get away from
>>> PF_RING as my hardware interface layer due to additional
>>> complications of out-of-kernel bolt-ons. I’ve really enjoyed the
>>> performance achieved using AF_PACKET for applications that support
>>> it. In Bro (and Suricata and now netsniff-ng), in order to support
>>> FANOUT_HASH mode provided by the kernel, the packets are spread across
>>> multiple processes. The problem is that these applications need symmetric
>>> hashing of the connections.
>>> The current ixgbe driver does not provide this guarantee in its RSS hashing.
>>> Recent changes in the kernel further made this a problem by accepting
>>> the hardware calculated hash, when available (for performance
>>> reasons, this would be great, unless you needed symmetric hashing).
>>>
>>> I ran across several discussions where an ixgbe driver was patched to
>>> support this, and also ran across this blog post [1] that used DPDK
>>> to override the driver hash. Can we just update the hash key in the
>>> mainline driver so that we get both good distribution of connections
>>> and symmetric hashing? It basically comes down to swapping out the
>>> hash key to ensure the first 32-bits (src ip) match the second
>>> 32-bits (dst ip) and the next 16-bits (src port) match the following
>>> 16-bits (dst port). A published paper provides a key change that
>>> modifies the standard Toeplitz hash function, which does not provide these
>>> guarantees.
>>>
>>> So… Can we just have a better hash function in hardware (via the
>>> driver) that feeds RSS (and now AF_PACKET)?
>>>
>>> Thanks!
>>>
>>> [1] http://galsagie.github.io/2015/02/26/dpdk-tips-1/
>>> <http://galsagie.github.io/2015/02/26/dpdk-tips-1/>
>>> —
>>> Derek Ditch
>>> de...@criticalstack.com
>>> GPG: 0x2543A3B5
>>
>
------------------------------------------------------------------------------
Attend Shape: An AT&T Tech Expo July 15-16. Meet us at AT&T Park in San
Francisco, CA to explore cutting-edge tech and listen to tech luminaries
present their vision of the future. This family event has something for
everyone, including kids. Get more information and register today.
http://sdm.link/attshape
_______________________________________________
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
