I have connected the AGX Xavier (flashed with the prebuild image) with aquantia 10G NIC card over PCIe connection. Also another AGX Xavier (flashed with the Android lineage OS) with customised 10G ethernet switch over PCIe connection.
Expecting a throughput of atleast 5G when doing the an TCP/UDP tranfer between the NIC Cards.
While doing the TCP client from the Linux AGX device, the throughput for the client itself is coming around 120Mbs (which is exceptionally low)
The command I issued is as below:
iperf3 -c 5.5.5.6 -b 10G -w 500k -l 32k
Thus the throughput at the receiver side of the android agx xavier is also less. Could you please help me to figure out the issue.
When I issue the UDP commands I can see an improvement in the throughput value (2.5Gbps) from the client side, but the android server is only accepting throughput around 1Gbps.
Could you please specify any specific configuration is required to improve the throughput of the ethernet switch.
I have the flashed both the AGX Xavier with the prebuilt images provided in the nvidia with and with applying the rt patches.
I am observing almost the same behaviour in all the situations.
I am attaching the obervations in both scenerio.
with_rt_patches.tar —> It contains the output of the TCP and UDP with the 1500 and 10K MTU values.
without_rt_patches.tar —> It contains the output of the TCP and UDP with the 1500 and 10K MTU values.
Please help to figure out what is the bottleneck to increase the throughput to atleast 5G for both the MTU conditions (1500 and 10K).
Later I can see a slight improvement in the throughput values for the 1500 MTU.
The throughput values are > 1.5Gbps, but less than 3Gbps for both the TCP and UDP modes.
For 10K MTU we are able to receive a throughput of 6Gbps, which is very well acceptable. Also for the 10K MTU the normal ipef3 command only is required.
iperf3 -c 5.5.5.5 -u -b 10G
Could you please let me know why the value is less for the 1500 MTU.
Hi,
We support Xavier with Jetpack 4 and 5 releases. Please try Jetpack 4.6.3 and 5.1.1. See if the issue occurs on either platform, or both. Jetpack 4 is in Kernel 4.9 and 5 is in Kernel 5.10.