[USRP-users] questions about uhd-dpdk with n310

Sam Reiter sam.reiter at ettus.com
Fri Nov 8 14:22:11 EST 2019


Panny Wang,

That's about the expected behavior currently. From the test benchmarks that
we have published, 3x3 @122.88 is not something that has been sustained
without some kind of flow control issue. If you want to explore building a
system to move towards this, I would recommend that you start with your
CPU. Processor clock speed is key in boosting the throughput of your host
machine, but 3x3 @122.88 isn't something I can guarantee you'll be able to
hit at this time.

Sam Reiter

On Thu, Nov 7, 2019 at 7:38 PM <ruoyilei at 126.com> wrote:

> Hello, Sam Reiter.
>
> Thank you for your reply.I have a test with benchmark, 2x2 @122.88M is
> ok(maybe the cpu “performance” is diabled in the initial test).I give the
> command with 3x3 at 122.88M. The output is in the attachment.
>
>      /usr/local/lib/uhd/examples/benchmark_rate  \
>
>     --args
> "type=n3xx,mgmt_addr=192.168.2.230,addr=192.168.10.2,second_addr=192.168.20.2,master_clock_rate=122.88e6,use_dpdk=1"
> \
>
>     --duration 60 \
>
>     --channels "0,1,2" \
>
>     --rx_rate 122.88e6 \
>
>     --rx_subdev "A:0 A:1 B:0" \
>
>     --tx_rate 122.88e6 \
>
>     --tx_subdev "A:0 A:1 B:0"
>
>
>
> Best,
>
> Panny Wang
>
>
>
> *发件人:* Sam Reiter <sam.reiter at ettus.com>
> *发送时间:* 2019年11月8日 4:41
> *收件人:* 王盼 <ruoyilei at 126.com>
> *抄送:* usrp-users at lists.ettus.com
> *主题:* Re: [USRP-users] questions about uhd-dpdk with n310
>
>
>
> Panny Wang,
>
>
>
> The cpufreq-info looks good, but the ifconfig at the bottom is a bit
> confusing with what you've sent over up to this point. Can you send the
> exact ./benchmark_rate command that you're using (with all args included)
> to produce the output you sent over initially? The MPMD info in the last
> couple doesn't seem consistent with this ifconfig output:
>
>
>
> enp7s0f0: flags=4163<UP,BROADCAST,RUNNING,MULTICAST>  mtu 1500
>         inet 192.168.1.73  netmask 255.255.255.0  broadcast 192.168.1.255
>         inet6 fe80::9604:9cff:fed2:b1a3  prefixlen 64  scopeid 0x20<link>
>         ether 94:04:9c:d2:b1:a3  txqueuelen 1000  (Ethernet)
>         RX packets 114457  bytes 8586410 (8.5 MB)
>         RX errors 0  dropped 3  overruns 0  frame 0
>         TX packets 179513  bytes 37029298 (37.0 MB)
>         TX errors 0  dropped 0 overruns 0  carrier 0  collisions 0
>         device memory 0x95e80000-95efffff
>
> enp7s0f1: flags=4163<UP,BROADCAST,RUNNING,MULTICAST>  mtu 8000
>         inet 192.168.2.254  netmask 255.255.255.0  broadcast 192.168.2.255
>         inet6 fe80::9604:9cff:fed2:b1a4  prefixlen 64  scopeid 0x20<link>
>         ether 94:04:9c:d2:b1:a4  txqueuelen 1000  (Ethernet)
>         RX packets 3404  bytes 296849 (296.8 KB)
>         RX errors 0  dropped 0  overruns 0  frame 0
>         TX packets 2196  bytes 243446 (243.4 KB)
>         TX errors 0  dropped 0 overruns 0  carrier 0  collisions 0
>         device memory 0x95e00000-95e7ffff
>
> lo: flags=73<UP,LOOPBACK,RUNNING>  mtu 65536
>         inet 127.0.0.1  netmask 255.0.0.0
>         inet6 ::1  prefixlen 128  scopeid 0x10<host>
>         loop  txqueuelen 1000  (Local Loopback)
>         RX packets 63270  bytes 4016936 (4.0 MB)
>         RX errors 0  dropped 0  overruns 0  frame 0
>         TX packets 63270  bytes 4016936 (4.0 MB)
>         TX errors 0  dropped 0 overruns 0  carrier 0  collisions 0
>
>
>
> Sam Reiter
>
>
>
> On Mon, Nov 4, 2019 at 7:32 PM 王盼 <ruoyilei at 126.com> wrote:
>
> Hey Sam Reiter,
>
> The output of "*cpufreq-info && ifconfig*" is in the attchment. At the
> same time I put more information about my system there.
>
> Both 10GbE links are binded to dpdk, so ifconfig can not output them.My
> cpu clock is 2.7GHz,maybe it is not powerfull enough.
>
> It would be great If you can help me confirm my configuration .
>
> *王盼*
>
> ruoyilei at 126.com
>
> 签名由 网易邮箱大师 <https://mail.163.com/dashi/dlpro.html?from=mail81> 定制
>
> On 11/5/2019 04:13,Sam Reiter<sam.reiter at ettus.com> <sam.reiter at ettus.com>
> wrote:
>
> Hey Panny Wang,
>
>
>
> You're correct, you should specify a second address with addr/second_addr,
> rather than addr0/addr1 - my bad. [1]
>
>
>
> Assuming you're using both 10GbE links correctly, my next step would be to
> investigate the processor you're using. Something with a higher clock speed
> is generally recommended for higher streaming rates.
>
>
>
> Would you be able to send over the output of:
>
>
>
> *cpufreq-info && ifconfig*
>
>
>
> Best,
>
>
>
> Sam Reiter
>
>
>
> [1]
> https://kb.ettus.com/Using_Dual_10_Gigabit_Ethernet_on_the_USRP_X300/X310
>
>
>
> On Sun, Nov 3, 2019 at 8:53 PM 王盼 <ruoyilei at 126.com> wrote:
>
> Hello,
>
> Sam Reiter. When leveraging dual 10GbE links,I  specify" addr=192.168.20.2,second_addr=192.168.10.2",last
> email I didn't give the example . The result is not much diffrent with
> use a single 10GbE link.
>
> I think it is  "addr=<xxx.xxx.xxx.xxx>,second_addr=<xxx.xxx.xxx.xxx>" but
> not "addr0=<xxx.xxx.xxx.xxx>,addr1=<xxx.xxx.xxx.xxx>". when
> use  "addr0=<xxx.xxx.xxx.xxx>,addr1=<xxx.xxx.xxx.xxx>",I get errors:
>
> *[INFO] [MPMD] Initializing 3 device(s) in parallel with args:
> mgmt_addr0=192.168.2.230,type0=n3xx,product0=n310,serial0=316645B,claimed0=False,mgmt_addr1=192.168.2.230,type1=n3xx,product1=n310,serial1=316645B,claimed1=False,mgmt_addr2=192.168.2.230,type2=n3xx,product2=n310,serial2=316645B,claimed2=False,type=n3xx,mgmt_addr=192.168.2.230,addr1=192.168.10.2,addr2=192.168.20.2,master_clock_rate=122.88e6,use_dpdk=1*
>
> *[ERROR] [RPC] Someone tried to claim this device again (From:
> 192.168.2.254)*
>
> *[WARNING] [MPM.RPCServer] Someone tried to claim this device again (From:
> 192.168.2.254)*
>
> *Error: RuntimeError: Error during RPC call to `claim'. Error message:
> Someone tried to claim this device again (From: 192.168.2.254)*
>
> *root at seu73:/home/seu# *
>
> On 11/2/2019 02:30,Sam Reiter<sam.reiter at ettus.com> <sam.reiter at ettus.com>
> wrote:
>
> Panny Wang,
>
>
>
> I notice that you're only specifying a single streaming address in your
> call to benchmark rate, implying that you're only leveraging a single 10GbE
> link. You can specify "addr0=<xxx.xxx.xxx.xxx>,addr1=<xxx.xxx.xxx.xxx>" in
> your device args.
>
>
>
> Best,
>
> Sam Reiter
> SDR Applications Engineer
> Ettus Research
>
>
>
> On Wed, Oct 30, 2019 at 3:20 AM 王盼 via USRP-users <
> usrp-users at lists.ettus.com> wrote:
>
> Hello,
>
>  Nate.I want to use DPDK in UHD with N310 follow
> https://files.ettus.com/manual/page_dpdk.html
> <https://files.ettus.com/manual/page_dpdk.html.>,but the result is not satisfactory.I
> got you have some research about this from the user-list emails(*With an
> i7-4790k / Intel x520-DA2 and N310, to stream at full duplex, over two
> channels at 125 MS/s, the lowest I can run my CPU clock freq at without
> flow control errors is 3.8 GHz using benchmark_rate and the native
> networking stack. Using DPDK I can run 2x2 @ 125 MS/s with my CPU freq
> locked at 1.5 GHz with no flow control errors. *).
>
> May be you can do me a favor and have some idea about my quesion.
>
> (1) I use benchmark_rate to test the streaming performance, I only got
> 122.88MS/s for 1channel, or 61.44MS/s for 2x2. run with 2x2 at 122.88MS/s ,
> a lot of samples dropped.
>
> Unfortuately, my destination is 4x4 at 122.88MS/s. I don't know is it
> possible for my present host machine, or what configuration of host machine
> should have?
>
> ubuntu server 18.04    uhd:3.14.1.1  dpdk 17.11.6   dual 10GbE links (XG
> image loaded)
>
> host machine: 4 node, 8 cores in each node, tota 32 cores, cpu: Intel(R)
> Xeon(R) CPU E5-4650 0 @ 2.70GHz
>
> more informations about my host machine is in the attchachment.(hypethread
> closed, cpufrequtils GOVERNOR="perfomance")
>
> *   --args
> "type=n3xx,mgmt_addr=192.168.1.104,addr=192.168.20.2,master_clock_rate=122.88e6,use_dpdk=1"
> \*
>
> *   --duration 60 \*
>
> *   --channels "0,1" \*
>
> *   --rx_rate 122.88e6 \*
>
> *   --rx_subdev "A:0 A:1" \*
>
> *   --tx_rate 122.88e6 \*
>
> *   --tx_subdev "A:0 A:1" *
>
> *   Benchmark rate summary:*
>
> *     Num received samples:     2744145668*
>
> *  Num dropped samples:      6030320380*
>
> *  Num overruns detected:    921*
>
> *  Num transmitted samples:  14684137560*
>
> *  Num sequence errors (Tx): 0*
>
> *  Num sequence errors (Rx): 0*
>
> *  Num underruns detected:   67231*
>
> *  Num late commands:        0*
>
> *  Num timeouts (Tx):        0*
>
> *  Num timeouts (Rx):        0*
>
>
> (2) In the  txrx_loopback_to_file test ,when I use the default --setting
> for 4*4channels, there is a error *UUUUError: Receiver error
> ERROR_CODE_LATE_COMMAND* .
>
> I change it to --setting 1 ,it works.
>
> I want to know the influence to my streaming or sample datas if 1
> increase  --setting?
>
> *(--settling arg (=0.20000000000000001) settling time (seconds)
> before receiving)*
>
>
>
> Much appreciated.
>
>
>
> Regards,
>
> Panny Wang
>
>
>
>
>
> _______________________________________________
> USRP-users mailing list
> USRP-users at lists.ettus.com
> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ettus.com/pipermail/usrp-users_lists.ettus.com/attachments/20191108/8cf634a0/attachment.html>


More information about the USRP-users mailing list