On 28/02/2023 18:56, Eugene Grayver wrote:
Hello,
I have a simple application (C++) that essentially copies input to
output. Using X310, direct connect 10 GbE, DPDK.
Each thread is tied to an isolated core, none of the cores (other than
DPDK) uses more than 75%.
UHD:
GNU C++ version 9.4.0; Boost_107100; DPDK_19.11; UHD_4.4.0.0-0-g5fac246b
Kernel command line:
|iommu=pt intel_iommu=on hugepages=1024 isolcpus=1-7 nohz_full=1-7
rcu_nocbs=1-7 irqaffinity=0 intel_idle.max_cstate=0 ipv6.disable=0
fips=1|.
I have turned off all the services I can think of, including irqbalance.
I am seeing an underflow event every ~33.4s. Very repeatable across
runs. Anyone have any idea what could be causing this?
Eugene.
This is a "loopback through the computer stack" type application?
Basically a receiver and a transmitter, in separate threads, but
sharing a buffer between them?
These can be tricky, and plenty of others have "stumbled" with this type
of app, particularly at high sample rates. I've never done one of
these myself, but I suspect that some kind of elastic FIFO mechanism
will be required.
I"m happy to have others chime in, but this class of application,
despite its *appearance* of simplicity can reveal the limitations of
ordinary general-purpose operating systems, and their schedulers, and
buffer-management systems.
_______________________________________________
USRP-users mailing list -- usrp-users@lists.ettus.com
To unsubscribe send an email to usrp-users-le...@lists.ettus.com
