Hide Forgot
Description of problem: When using tc netem to emulate a high bandwidth / high delay connection the TCP throughput is about 50x slower when you turn on variable delay. Version-Release number of selected component (if applicable): 2.6.32 How reproducible: Very. Steps to Reproduce: 1. Test bed: client <-GigE-> linix bridge w/ tc netem <-GigE-> HTTP/FTP server 2. On linux bridge do tc qdisc add dev eth0 root handle 1:0 netem delay 25ms tc qdisc add dev eth1 root handle 2:0 netem delay 25ms Verify RTT with ping - should be about 50.2 Test throughput with wget 3. On linux bridge do tc qdisc change dev eth0 root handle 1:0 netem delay 25ms 0.1ms distribution normal tc qdisc change dev eth1 root handle 2:0 netem delay 25ms 0.1ms distribution normal Verify RTT with ping - should be between 49ms - 51ms Retest throughput with wget Actual results: Throughput drops dramatically Expected results: Throughput should drop much less than 50%, not 50x Additional info:
Since RHEL 6.2 External Beta has begun, and this bug remains unresolved, it has been rejected as it is not proposed as exception or blocker. Red Hat invites you to ask your support representative to propose this request, if appropriate and relevant, in the next release of Red Hat Enterprise Linux.
Can you please try to reproduce this bug with 6.3 Beta. We have fixed some issues around this area, and we believe it should be fixed now. If not please send new logs. If you cannot try 6.3Beta for some other restrictions then please let us know as well. Thanks
This bug looks to be fixed. I used ttcp between two RHEL systems with a RHEL 6.3beta based bridge in between them. I get the following two results with variable delay turned off then turned on. Test 1: bridge settings on eth0 and eth1: tc qdisc add dev eth0 root netem delay 25ms tc qdisc add dev eth1 root netem delay 25ms result: ttcp-r: 536870912 bytes in 17.675 real seconds = 28.968 MB/sec +++ Test 2: bridge settings on eth0 and eth1: tc qdisc add dev eth0 root netem delay 25ms 1us distribution normal tc qdisc add dev eth1 root netem delay 25ms 1us distribution normal result: ttcp-r: 536870912 bytes in 21.108 real seconds = 24.256 MB/sec +++ So ~29MB/sec versus ~24MB/sec. Much, much better.
Thanks for letting us know Peter. I will close this bug. Thanks