Regression testing done on an older/previously supported adapter, SFC
7000 series.
The netperf suite of TCP/UDP STREAM and RR, and TCP_RR ran for ~2 days,
with results in the same ballpark as the original kernel and test kernels.
Now waiting for test results with the new/requested adapter
before marking verification done/successful.
Summary: test name, mtu sizes, original/test/proposed kernel results.
TCP_CRR
1500/1500
ORIG 4550-4560
TEST 4550-4580
PROP 5260-5316
9000/9000
ORIG 4557
TEST 4570
PROP 5260-5300
TCP_RR
1500/1500
ORIG 32531
TEST ~31k,32k
PROP 32180-34277
9000/9000
ORIG 31620
TEST 27k-30k-36k
PROP 27k-33k-34k
TCP_STREAM
1500/1500
ORIG 9406
TEST 9403
PROP 9405
9000/9000
ORIG 9883
TEST 9887
PROP 9887
UDP_RR
1500/1500
ORIG ~36k/~37k
TEST ~36k/~37k
PROP ~36k
9000/9000
ORIG ~35k/~37k
TEST 33k-37k
PROP ~35.8k/~36.6k
UDP_STREAM
1500/1500
ORIG 8.6k/8.9k
TEST 8.9k
PROP 8.6k/8.7k
9000/9000
ORIG 8.7k
TEST 8.7k/8.8k
PROP 8.7k
--
You received this bug notification because you are a member of Kernel
Packages, which is subscribed to linux in Ubuntu.
https://bugs.launchpad.net/bugs/1836635
Title:
Bionic: support for Solarflare X2542 network adapter (sfc driver)
Status in linux package in Ubuntu:
Invalid
Status in linux source package in Bionic:
Fix Committed
Status in linux source package in Cosmic:
Invalid
Status in linux source package in Disco:
Invalid
Status in linux source package in Eoan:
Invalid
Bug description:
[Impact]
* Support for Solarflare X2542 network adapter
(Medford2 / SFC9250) in the Bionic sfc driver.
* This network adapter is present on recent hardware,
at least HP 2019 and Dell PowerEdge R740xd systems.
* On recent-hardware deployments that would rather use
the Bionic LTS / GA supported kernel and cannot move
to HWE kernels this adapter is non functional at all.
[Test Case]
* The X2542 adapter has been exercised with iperf3 and nc
across 2 hosts on 25G link speed w/ MTUs 1400/1500/9000
on both directions, for 1 week.
Its performance is on par with the Cosmic 4.18 kernel
(which contains all these patches) and the out-of-tree
driver from the vendor.
* The 7000 series adapter (for regression testing an old model,
supported previously) has been exercised with iperf and netperf
(TCP_STREAM, UDP_STREAM, TCP_RR, UDP_RR, and TCP_CRR) in one
host (client/server in different adapter ports isolated with
network namespaces, so traffic goes through the network switch),
on 10G link speed on MTUs 1500/9000, for 1 weekend.
No regressions observed between the original and test kernels.
[Regression Potential]
* The patchset touches a lot of the sfc driver, so the potential
for regression definitely exists. Thus, a lot of consideration
and testing happened:
* It has been tested on other adapter which uses the old code,
and no regressions were found so far (see 7000 series above).
* The patchset is exclusively cherry-picks, no single backport.
* The patchset essentially moves the Bionic driver up in the
upstream 'git log --oneline -- drivers/net/ethernet/sfc/':
- since commit d4a7a8893d4c ("sfc: pass valid pointers from
efx_enqueue_unwind")
- until commit 7f61e6c6279b ("sfc: support FEC configuration through
ethtool")
- except for 2 commits (not needed / unrelated)
- commit 42356d9a137b ("sfc: support RSS spreading of ethtool ntuple
filters")
- commit 9baeb5eb1f83 ("sfc: falcon: remove duplicated bit-wise or of
LOOPBACK_SGMII")
- plus 2 more recent commits (fixes)
- commit 458bd99e4974 ("sfc: remove ctpio_dmabuf_start from stats")
- commit 0c235113b3c4 ("sfc: stop the TX queue before pushing new
buffers")
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1836635/+subscriptions
--
Mailing list: https://launchpad.net/~kernel-packages
Post to : [email protected]
Unsubscribe : https://launchpad.net/~kernel-packages
More help : https://help.launchpad.net/ListHelp
