[srslte-users] ZMQ performance issues?

[Andre Puschmann]

Hi David,

On 29/6/20 20:45, David Rupprecht wrote:
> I have just tested the ZMQ testing environment. Particularly, I run the
> ./test/run_lte.sh script with the parameters "../build_release/ 25"
> (with sudo) on my Ubuntu18.04 with a i7-7820. I'm on the latest commit
> (master) and build it srsLTE in release mode. However, while running the
> script I got some performance and general errors. To understand the
> problems and I got some general questions to check my configuration. I
> also attached the logs.

I am afraid but in the current version of the "run_lte.sh" script in
20.04.1 there are still a few issues when it comes to timing between PHY
and upper layer threads. Those are all fixed and will be made available
in the upcoming release.

> 
> First of all, the script detects multiple PRACHs: "Error. Detected 5
> PRACH(s). But should be only 1." While in this test run only have 5
> PRACH(s) occured, other test runs contain up to 10 PRACH(s). Should it
> be exact one PRACH? The eNodeB detects 4 times : [RRC ] [W] Activity
> timer for rnti=0x46 expired after 5000 ms. However, there is a sleep
> between ping and iperf which should trigger the inactivity timer.
> Nevertheless, there shall be only 2 times triggered. What many PRACH
> does occur
> 
> Further, the UE has some problems with the detects some slow task "[W]
> Detected slow task processing". When I run the srsUE with a USRP this
> does not occur. Is this a known problem when using ZMQ or srsUE and
> srsENB on the same machine? Do you see similar warnings with the
> run_lte.sh script?

Yes. Those warnings are expected. Also when you run with address
sanitizer or valgrind the processing can take a bit.
For ZMQ, you can safely ignore them, start with
"--stack.have_tti_time_stats false" so they are not displayed. They are
enabled by default because on real HW this flag can indicate if the
upper layer processing takes too long.

> 
> Further, the 165 Kbits/sec seems a bit slow for a 25PRB setup. Are those
> parameters in the normal range?

We don't do full rate tests in the run_lte.sh script. So its expected to
not achieve the max throughput for any of the configs, yes.


Cheers
Andre