On 2/3/20 9:26 PM, Ed Greshko wrote:
On 2020-02-04 11:11, Dave Ulrick wrote:
I am experiencing just one minor issue having to do with throughput on the PC's built-in Ethernet interface:
Sounds as if it may be this?
https://bugzilla.redhat.com/show_bug.cgi?id=1797232
I think I've pinpointed a possible cause for my Ethernet slowdown.
Evidently the Ethernet chipset on the affected PC has some bugs insofar
as offload functionality and some relatively recent patches to the r8169
kernel module have had a negative effect on performance. The crucial
factor seems to be the tcp-segmentation-offload parameter that's
viewable with ethtool -k.
The PC with slow transmit speeds has tcp-segmentation-offload turned off:
tcp-segmentation-offload: off
and 'ethtool' won't let me turn it on:
# ethtool -K enp3s0 tso on
Cannot change tcp-segmentation-offload
Could not change any device features
Another PC which doesn't suffer from the transmit slowness has
tcp-segmentation-offload enabled:
tcp-segmentation-offload: on
The first PC gets an iperf3 bitrate of ~ 870 Mbits/sec versus the second
PC's bitrate of ~ 940 Mbits/sec.
If I disable tcp-segmentation-offload on the second PC:
# ethtool -K enp4s0 tso off
and run iperf3, I get ~ 880 Mbits/sec, very similar to the throughput on
the slower PC. If I turn tso back on, it goes back to ~ 940 Mbits/sec.
This makes it look as though having tcp-segmentation-offload disabled
might be the reason why the slower PC is getting slower transmit throughput.
Maybe a recent Linux kernel distributed an r8169 module that disabled
tcp-segmentation-offload for:
r8169 0000:03:00.0 eth0: RTL8168evl/8111evl, 94:de:80:21:61:12, XID 2c9,
IRQ 30
but let it continue to work OK for:
r8169 0000:04:00.0 eth0: RTL8168d/8111d, 00:30:67:6a:e2:64, XID 281, IRQ 35
This looks very similar to my issue--RTL8168evl chipset &
tcp-segmentation-offload--excepting that they're discussing kernel 4.19
whereas I'm using 5.4.17:
https://lore.kernel.org/netdev/217e3fa9-7782-08c7-1f2b-8dabacaa83f9@xxxxxxxxx/T/
I've ordered a PCIE Gigabit Ethernet card in hopes that it will have a
chipset that isn't affected by this issue.
Dave
_______________________________________________
users mailing list -- users@xxxxxxxxxxxxxxxxxxxxxxx
To unsubscribe send an email to users-leave@xxxxxxxxxxxxxxxxxxxxxxx
Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: https://lists.fedoraproject.org/archives/list/users@xxxxxxxxxxxxxxxxxxxxxxx
