[srsran-users] Performance issues

Wed Oct 4 12:57:38 UTC 2023

It's seems the MTU of your interface is very small, you need to increase it
to 9000. You can do it with ifconfig command.

On Wed, 4 Oct 2023 at 14:53, <08.chancy_fantasy at icloud.com> wrote:

> Hello,
>
> Just realised I have not CC’d srsran-users in my last reply to Ismael. The
> files are attached here although I am not sure what is the best way to
> share logs and other files with the members.
>
> @Ismael, one further development since yesterday - The following warnings
> from UHD disappeared after I upgraded my USRP x300 FPGA images:
>
> *[WARNING] [X300] For the 192.168.40.2 connection, UHD recommends a send
> frame size of at least 8000 for best*
> *performance, but your configuration will only allow 2664.This may
> negatively impact your maximum achievable sample rate.*
> *Check the MTU on the interface and/or the send_frame_size argument.*
> *[WARNING] [X300] For the 192.168.40.2 connection, UHD recommends a
> receive frame size of at least 8000 for best*
> *performance, but your configuration will only allow 2664.This may
> negatively impact your maximum achievable sample rate.*
> *Check the MTU on the interface and/or the recv_frame_size argument.*
>
> HOWEVER the behaviour remained the same afterwards - that is, the
> throughput and instability issues continue even with send/receive frames
> sizes being 8000.
>
> Kind Regards,
> .Nemo
>
>
>
> On 3 Oct 2023, at 15:29, muhasard77 at gmail.com wrote:
>
> Hi Ismael,
>
> Thanks for your prompt response.
>
> Please find attached the requested files.
>
> There are some warnings generated by the UHD driver that I was already
> trying to figure out without success so far.
> But, like you said if there is any other parameter tuning that I have got
> wrong, please highlight that too.
>
> I was running an iperf3 test from the same machine that was running the
> SDR code using the command shown below. You will see that the traffic
> reported by iperf3 vs SDR UE trace also do not match.
>
> You mentioned 100MHz SISO or 50MHz MIMO should not have much problem. Does
> that mean for SISO case, srslte can support 5 carrier components of 20MHz
> each and hence the maximum throughput that can be supported without any
> accelerators is 250Mbps?
>
> Thank you very much for your support!
>
> Kind Regards,
> .Nemo
>
>
> $ i*perf3 -c 10.79.4.20 -R -l 1400 -u -b 50m -t 100 -p 5201*
> Connecting to host 10.79.4.20, port 5201
> Reverse mode, remote host 10.79.4.20 is sending
> [  4] local 10.79.4.34 port 51679 connected to 10.79.4.20 port 5201
> [ ID] Interval           Transfer     Bandwidth       Jitter    Lost/Total
> Datagrams
> [  4]   0.00-1.00   sec  4.78 MBytes  40.1 Mbits/sec  0.294 ms  0/3583
> (0%)
> [  4]   1.00-2.00   sec  8.69 MBytes  72.9 Mbits/sec  0.200 ms  0/6508
> (0%)
> [  4]   2.00-3.00   sec  3.95 MBytes  33.1 Mbits/sec  0.364 ms  0/2955
> (0%)
> [  4]   3.00-4.00   sec  1.71 MBytes  14.3 Mbits/sec  0.462 ms  0/1279
> (0%)
> [  4]   4.00-5.00   sec  2.01 MBytes  16.9 Mbits/sec  0.456 ms  0/1506
> (0%)
> [  4]   5.00-6.00   sec  7.07 MBytes  59.3 Mbits/sec  0.218 ms  0/5296
> (0%)
> [  4]   6.00-7.00   sec  7.62 MBytes  63.9 Mbits/sec  0.855 ms  0/5709
> (0%)
> [  4]   7.00-8.00   sec  6.39 MBytes  53.6 Mbits/sec  0.205 ms  0/4786
> (0%)
> [  4]   8.00-9.00   sec  0.00 Bytes  0.00 bits/sec  0.205 ms  0/0 (0%)
> [  4]   9.00-10.00  sec  0.00 Bytes  0.00 bits/sec  0.205 ms  0/0 (0%)
> [  4]  10.00-11.00  sec  0.00 Bytes  0.00 bits/sec  0.205 ms  0/0 (0%)
> [  4]  11.00-12.00  sec  0.00 Bytes  0.00 bits/sec  0.205 ms  0/0 (0%)
> [  4]  12.00-13.00  sec  4.73 MBytes  39.7 Mbits/sec  1.074 ms  0/3544
> (0%)
> [  4]  13.00-14.00  sec  9.69 MBytes  81.2 Mbits/sec  0.490 ms  0/7254
> (0%)
> [  4]  14.00-15.00  sec  8.07 MBytes  67.7 Mbits/sec  0.344 ms  0/6048
> (0%)
> [  4]  15.00-16.00  sec  9.37 MBytes  78.6 Mbits/sec  0.244 ms  0/7015
> (0%)
> [  4]  16.00-17.00  sec  2.75 MBytes  23.0 Mbits/sec  0.277 ms  0/2056
> (0%)
> [  4]  17.00-18.00  sec  2.75 MBytes  23.1 Mbits/sec  0.419 ms  0/2060
> (0%)
> [  4]  18.00-19.00  sec  2.69 MBytes  22.5 Mbits/sec  0.696 ms  0/2012
> (0%)
> [  4]  19.00-20.00  sec  2.35 MBytes  19.7 Mbits/sec  0.238 ms  0/1760
> (0%)
> [  4]  20.00-21.00  sec  2.53 MBytes  21.2 Mbits/sec  1.330 ms  0/1894
> (0%)
> [  4]  21.00-22.00  sec  2.07 MBytes  17.4 Mbits/sec  1.786 ms  0/1553
> (0%)
> [  4]  22.00-23.00  sec  2.50 MBytes  20.9 Mbits/sec  1.223 ms  0/1869
> (0%)
>
>
> On 3 Oct 2023, at 13:42, Ismael Gomez <ismael.gomez at srs.io> wrote:
>
> Hi Nemo,
>
> On Tue, 3 Oct 2023 at 10:54, <08.chancy_fantasy at icloud.com> wrote:
>
>> Hello,
>>
>> I am relatively new to srslte (but not to radios and SDRs..)
>>
>> Already spent a couple of hours browsing / experimenting with the code
>> and I must say it is very clean and well-written code.
>> However, it seems that performance remains the number 1 limitation.
>> Ramping up to 75 PRBs on a Xeon 2.8GHz 8-core CPU, I notice a high degree
>> of instability running an iperf3 UDP DL maximum throughput test. Of course,
>> the throughput degrades sharply as the instability kicks-in (usually
>> triggered by an innocent looking couple of underrun packets or an RLF).
>> Note that these tests were run under controlled radio conditions with a
>> commercial-grade eNB / ePC.
>>
>> So, my first question is: can someone confirm the performance that I am
>> seeing? Or is it just down to my not-so-capable platform? Is there a
>> reference platform that can hit or approach maximum performance for say, 75
>> PRBs (100Mbps DL / 30 Mbps UL)?
>>
>
> We probably need to tune a few parameters there. Can you share your config
> files, console trace and log in warning please?
>
>
>>
>> Secondly, is it possible to off-load some of the heavy-lifting (turbo
>> decoding e.g.) to an external device? We have an FPGA based platform that
>> connects to the host via a PCI interface. Ignoring the development effort,
>> my concern here is whether the cost of such off-loading (over much slower
>> PCI bus and back) could already surpass the ~2ms processing window
>> available..
>>
>
> We provide support for the bbdev interface to use the Intel ACC100
> accelerator. But for those BW it's really not needed. Your Xeon 8-core
> should be able to do 100 MHz SISO or 50 MHz MIMO without major problems
> (assuming everything is optimally tuned)
>
>
>>
>> I have already combed through mailing list archives but could not find a
>> straight answer to the above questions.
>>
>> Kind Regards,
>> .Nemo
>> _______________________________________________
>> srsran-users mailing list
>> srsran-users at lists.srsran.com
>> https://lists.srsran.com/mailman/listinfo/srsran-users
>
>
>
> _______________________________________________
> srsran-users mailing list
> srsran-users at lists.srsran.com
> https://lists.srsran.com/mailman/listinfo/srsran-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.srsran.com/pipermail/srsran-users/attachments/20231004/559fa0f4/attachment.htm>