Hide Forgot
Description of problem: performance regression of tcp_stream throughput between kernel-2.6.18-259 and kernel-2.6.18-260.While testing,I disabled tso and enable gso on host and external host. kernel-version Throughput 10^6bits/sec (TCP_STREAM) 2.6.18-260.el5 8677.11 2.6.18-259.el5 2048.93 Version-Release number of selected component (if applicable): kernel-2.6.18-260.el5 kernel-2.6.18-259.el5 How reproducible: 100% Steps to Reproduce: 1.run netserver on rhel 5.7 host with kernel-2.6.18-259.el5. 2.run netperf on ex-host ./netperf -H 192.168.0.11 -l 60 TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.0.11 (192.168.0.11) port 0 AF_INET Recv Send Send Socket Socket Message Elapsed Size Size Size Time Throughput bytes bytes bytes secs. 10^6bits/sec 87380 16384 16384 60.00 8677.11 3.switching to kernel-2.6.18-260.el5 on host. 4.run netperf again on ex-host ./netperf -H 192.168.0.11 -l 60 TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 192.168.0.11 (192.168.0.11) port 0 AF_INET Recv Send Send Socket Socket Message Elapsed Size Size Size Time Throughput bytes bytes bytes secs. 10^6bits/sec 87380 16384 16384 60.01 2048.93 Actual results: Expected results: Additional info: ethtool -k eth2 Offload parameters for eth2: Cannot get device udp large send offload settings: Operation not supported rx-checksumming: on tx-checksumming: on scatter-gather: on tcp segmentation offload: off udp fragmentation offload: off generic segmentation offload: on generic-receive-offload: on ethtool eth2 Settings for eth2: Supported ports: [ FIBRE ] Supported link modes: 1000baseT/Full 10000baseT/Full Supports auto-negotiation: Yes Advertised link modes: 1000baseT/Full 10000baseT/Full Advertised auto-negotiation: Yes Speed: 10000Mb/s Duplex: Full Port: FIBRE PHYAD: 0 Transceiver: external Auto-negotiation: on Supports Wake-on: d Wake-on: d Current message level: 0x00000007 (7) Link detected: yes ethtool -i eth2 driver: ixgbe version: 3.2.9-k2 firmware-version: 0.9-3 bus-info: 0000:0f:00.0
There are no ixgbe changes between -259 and -260, so I do not suspect this is specifically a problem on ixgbe. What other 10GbE interfaces were tested and did not show the issue?
BTW, we are also tracking what may or may not be an ixgbe issue in bug 706757.
(In reply to comment #4) > There are no ixgbe changes between -259 and -260, so I do not suspect this is > specifically a problem on ixgbe. What other 10GbE interfaces were tested and > did not show the issue? As -260 was removed from brew,I do the comparison again between -259 and -266. Keep the netperf server with -259 kernel ,switching netperf sender with kernel from -259 to -266.The results show there is no regression now. netperf ci : ./netperf -H 192.168.0.11 -l 60 (with default message size :16384 bytes) Sender Receiver | -259 | -266 | ----------------+----------------+---------+-----------+ tso on gso off rsc on gro on 8700.17 8716.22 ----------------+----------------+---------+-----------+ tso on gso off rsc on gro off 8804.30 8767.13 ----------------+----------------+---------+-----------+ tso off gso on rsc on gro on 7494.25 7582.54 ----------------+----------------+---------+-----------+ tso on gso off rsc off gro on 6657.81 6627.00 ----------------+----------------+---------+-----------+ tso on gso off rsc off gro off 6258.57 6591.09 ----------------+----------------+---------+-----------+ tso off gso on rsc off gro on 6218.01 6568.12 ----------------+----------------+---------+-----------+
Closing as NOTABUG based on results in comment #7.