I should add that when I said "I have not done this", I meant with N321s in
this proposed configuration.  However, I am presently using an XL710 (QDA1)
with a QSFP+ optical transceiver and MTP/LC breakout fiber to connect to
two N310s (using SFP+ optical transceivers).  I have done this both with
single-mode and multimode fiber.
Rob

On Thu, Feb 4, 2021 at 9:32 AM Rob Kossler <rkoss...@nd.edu> wrote:

> Hi Pat,
> Are you using fiber or copper to connect?  If you are using fiber, it
> seems to me it could work with a three QSFP optical transceivers and three
> MTP/LC breakout fibers that you would connect via couplers such that Host
> Lanes 0 and 1 go to the first N321 Lanes 0 and 1 while the Host Lanes 2 and
> 3 go to the second N321 Lanes 0 and 1.  The host NIC would be configured
> 1x4.  Note that I have not done this, but I am planning to purchase N321s
> soon and so I am interested in this topic.
> Rob
>
> On Wed, Feb 3, 2021 at 5:34 PM Patrick Kane via USRP-users <
> usrp-users@lists.ettus.com> wrote:
>
>> I still cannot use a single 2-port QSFP+ NIC to connect to 2x  N321s.
>> Using the Intel tools to set the NIC to 2x2x10, the NIC doesn't recognize
>> the second physical port as a valid connection to a USRP, but works as a
>> loopback or a connection to another NIC. I asked Intel regarding the issues
>> to see if we were configuring the NIC incorrectly, and this is the response
>> we got:
>>
>> *Response from Intel on XL710-QDA2*
>>
>> “When configured for 2x2x10 the lanes are split between the ports. The A
>> configuration utilizes lanes 1 and 2 on the top port and lanes 3 and 4 on
>> the bottom port. These correlate to the physical connections on the QSFP+
>> module. If you are able to configure the port of the other device to
>> utilize lanes 3 and 4, then it should be able to connect to the bottom port
>> in the A configuration.”
>>
>>
>>
>> *Problem:* XQ FPGA image doesn’t accept connections from lanes 2 and 3,
>> just 0 and 1. This prevents the bottom port on the Intel NIC from working
>> in the 2x2x10 configuration:
>>
>>
>> InterfaceHGXGWXXQAQ
>> SFP+ 0 1 GbE 10 GbE White Rabbit White Rabbit 10 GbE
>> SFP+ 1 10 GbE 10 GbE 10 GbE Unused 10 GbE
>> QSFP+ lane 0 Unused Unused Unused 10 GbE Aurora
>> QSFP+ lane 1 Unused Unused Unused 10 GbE Aurora
>> QSFP+ lane 2 Unused Unused Unused *Unused* Aurora
>> QSFP+ lane 3 Unused Unused Unused *Unused* Aurora
>> In the 2 N321 configuration, lane 2 and 3 are valid, but point to a
>> different USRP. I'm hoping theres a UHD or FW change that can prevent the
>> QSFP->2x10G breakout cable because that defeats the purpose of using a
>> QSFP+ NIC instead of a 4-port 10G NIC (Intel X710-DA4)
>>
>> Thanks,
>> Pat
>>
>> On Tue, Nov 24, 2020 at 1:01 PM Michael Dickens <
>> michael.dick...@ettus.com> wrote:
>>
>>> Hi Pat - I'm glad that info helped!
>>>
>>> Yes, I plan on adding this information into the N32x Getting Started
>>> Guide, once I have a better handle on it. Right now I have just a few data
>>> points & those are not consistent! and I don't know why! Thus ...
>>>
>>> Which Intel QSFP+ utility did you end up using? There are 2 that I can
>>> find:
>>>
>>> 1) EPCT:
>>> https://downloadcenter.intel.com/download/28933/Intel-Ethernet-Port-Configuration-Tool-Linux-
>>>
>>> This is the newer version that seems to work.
>>>
>>> 2) QCU :
>>> https://downloadcenter.intel.com/download/25851/Intel-QSFP-Configuration-Utility-Linux-Final-Release?product=46828
>>>
>>> This one is deprecated, though it still works to some extent.
>>>
>>> ===
>>>
>>> When I execute (A), I get the following options: "4x10" and "2x2x10". I
>>> do not get an "A" or "B" or "LOM" or whatever. Just literally those 2
>>> options.
>>>
>>> I think the first one means "1x(4x10)", meaning that just port0 is
>>> active & provides 4 data lates. I was hoping this option would work with a
>>> 1:4 SFP+ breakout cable from FS.com, but to the best of my testing I can
>>> get just 1 of those 4 SFP+ links to work. Supposedly if one uses the Intel
>>> 1:4 breakout cable this will work ... but that's paying $350 for an
>>> otherwise $50 cable! I'm still investigating here. Ideally this NIC would
>>> provide "2x(4x10)" with 2 1:4 breakout cables, to get double the SFP+
>>> density of current NICs (e.g., the X710-DA4 ort X722-DA4).
>>>
>>> The second one implies to me that both ports are available & providing 2
>>> data lanes each. The best I've been able to do is use "2x2x10" with port0;
>>> port1 doesn't seem to be working in this setting.
>>>
>>> Admittedly, I might need to update to the current Intel Linux drivers
>>> for the XL710 NIC. I usually let the OS handle this for me -- in this case,
>>> Ubuntu 20.04 latest. There are new Intel drivers from early November 2020,
>>> but I don't think the XL710 had any updates from the prior version.
>>>
>>> I'm curious what driver version & OS / version you're using ... maybe
>>> let's catch up off-list for a bit & see what we can figure out here.
>>> Cheers! - MLD
>>>
>>> On Tue, Nov 24, 2020 at 9:06 AM Patrick Kane <prkan...@gmail.com> wrote:
>>>
>>>> Hi Mike,
>>>>
>>>> That seemed to do the trick, thanks for info! At some point, can we
>>>> make these steps part of the N32x getting started docs?
>>>>
>>>> Also- the config utilty makes me choose 2x2x10 A, B, or LOM. Choosing A
>>>> disables the second port on the QDA2, and B disables the first port. LOM
>>>> disables both ports (expected because it's not a MB NIC). My ideal case is
>>>> using 2x N321s over QSFP on the same XL710-QDA2 NIC. Have you had any luck
>>>> in this configuration?
>>>>
>>>> Thanks,
>>>> Pat
>>>>
>>>> On Mon, Nov 23, 2020 at 9:23 PM Michael Dickens <
>>>> michael.dick...@ettus.com> wrote:
>>>>
>>>>> Hi Pat - I recently verified that the N321 QAFP+ interface works with
>>>>> UHD 4.0 release. I am also using an Intel XL710 (QDA2, but that probably
>>>>> doesn’t matter too much). The trick for me was using the Intel QSFP+ NIC
>>>>> configuration tool to set the NIC to 2x(2x10 Gb) mode. This is the setting
>>>>> that the N321 requires, and one that the NIC provides. Once that was set
>>>>> then you need to configures the host and USRP network interfaces as
>>>>> you normally would. After all of that, the link worked very nicely! I hope
>>>>> this is useful! - MLD
>>>>>
>>>>> On Nov 23, 2020, at 4:44 PM, Patrick Kane via USRP-users <
>>>>> usrp-users@lists.ettus.com> wrote:
>>>>>
>>>>> 
>>>>>
>>>>> I have an N321 connected to serial console and QSFP+ through a XL710
>>>>> Intel NIC. With the default HG image, I can connect through 1G and serial
>>>>> as expected. I updated the filesystem to UHD 4.0.0.0 using mender, and the
>>>>> build artifact reflects that this was successful. Then, after loading the
>>>>> XQ image (using 2x 10Gb lanes through QSFP+ port), I lose all ethernet
>>>>> connectivity through the 1G port SFP0 (expected), but I get the following
>>>>> output in the console window:
>>>>>
>>>>>
>>>>> [  451.560674] nixge 40000000.ethernet sfp0: Link is Up - 10Gbps/Full
>>>>> - flow control off
>>>>>
>>>>> [  453.800673] nixge 40000000.ethernet sfp0: Link is Down
>>>>>
>>>>> [  454.920676] nixge 40000000.ethernet sfp0: Link is Up - 10Gbps/Full
>>>>> - flow control off
>>>>>
>>>>> [  458.280672] nixge 40000000.ethernet sfp0: Link is Down
>>>>>
>>>>> [  459.400677] nixge 40000000.ethernet sfp0: Link is Up - 10Gbps/Full
>>>>> - flow control off
>>>>>
>>>>> [  462.760705] nixge 40000000.ethernet sfp0: Link is Down
>>>>>
>>>>> [  463.880678] nixge 40000000.ethernet sfp0: Link is Up - 10Gbps/Full
>>>>> - flow control off
>>>>>
>>>>> [  466.120673] nixge 40000000.ethernet sfp0: Link is Down
>>>>>
>>>>>
>>>>> uhd_usrp_probe:
>>>>>
>>>>>   _____________________________________________________
>>>>>
>>>>>  /
>>>>>
>>>>> |       Device: N300-Series Device
>>>>>
>>>>> |     _____________________________________________________
>>>>>
>>>>> |    /
>>>>>
>>>>> |   |       Mboard: ni-n3xx-31E00AC
>>>>>
>>>>> |   |   dboard_0_pid: 338
>>>>>
>>>>> |   |   dboard_0_serial: 31DB406
>>>>>
>>>>> |   |   dboard_1_pid: 338
>>>>>
>>>>> |   |   dboard_1_serial: 31DB407
>>>>>
>>>>> |   |   eeprom_version: 3
>>>>>
>>>>> |   |   fs_version: 20200914000806
>>>>>
>>>>> |   |   mender_artifact: v4.0.0.0_n3xx
>>>>>
>>>>> |   |   mpm_sw_version: 4.0.0.0-g90ce6062
>>>>>
>>>>> |   |   pid: 16962
>>>>>
>>>>> |   |   product: n320
>>>>>
>>>>> |   |   rev: 7
>>>>>
>>>>> |   |   rpc_connection: local
>>>>>
>>>>> |   |   serial: 31E00AC
>>>>>
>>>>> |   |   type: n3xx
>>>>>
>>>>> |   |   MPM Version: 3.0
>>>>>
>>>>> |   |   FPGA Version: 8.0
>>>>>
>>>>> |   |   FPGA git hash: be53058.clean
>>>>>
>>>>> |   |
>>>>>
>>>>> |   |   Time sources:  internal, external, gpsdo, sfp0
>>>>>
>>>>> |   |   Clock sources: external, internal, gpsdo
>>>>>
>>>>> |   |   Sensors: ref_locked, gps_locked, temp, fan, gps_gpgga,
>>>>> gps_sky, gps_time, gps_tpv
>>>>>
>>>>>
>>>>> Are there any configuration items needed to connect to the N321
>>>>> through the QSFP+ port. Note that I only see eth0, sfp0, sfp1, and int0 in
>>>>> /etc/network/interfaces.
>>>>>
>>>>>
>>>>> Thanks,
>>>>>
>>>>> Pat
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> USRP-users mailing list
>>>>> USRP-users@lists.ettus.com
>>>>> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com
>>>>>
>>>>> _______________________________________________
>> USRP-users mailing list
>> USRP-users@lists.ettus.com
>> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com
>>
>
_______________________________________________
USRP-users mailing list
USRP-users@lists.ettus.com
http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com

Reply via email to