problems with RED element ?
Brecht Vermeulen
brecht.vermeulen at rug.ac.be
Thu Jul 12 22:42:17 EDT 2001
Hi,
I'm doing some tests with the red element today and I've noticed some
strange things :
test configuration : make-ip-conf script but with a RED element before
each queue (as push)
parameters of the RED element (5,20,0.1 as described in
http://www.aciri.org/floyd/REDparameters.txt)
queue length is 25.
The router has 4 interfaces and there are sent 2 streams, one from
eth1->eth2 and one from eth3->eth4 (eth0 is a fifth control interface).
These streams are gradually increasing in bandwidth: 10Mbit, 20, ...
Packet size is 256 bytes.
* first problem : if you look at next results, for the 90 Mbit and 100
Mbit case (so 180 and 200 Mbit/s in total) one stream has a very high
drop rate while the other one has a zero drop rate. All dropped packets
are dropped in the red elements (seen in the handlers). The queue itself
only drops about 30packets.
Packet Bandwidth Sent Received Lost Lost (%)
size per flow Packets
256 10.00000 45,289 43,829 1,460 3.22374
256 10.00000 45,289 45,024 265 0.58513
256 20.00000 90,579 87,673 2,906 3.20825
256 20.00000 90,579 90,091 488 0.53876
256 30.00000 135,869 133,322 2,547 1.87460
256 30.00000 135,869 135,869 0 0.00000
256 40.00000 181,159 180,256 903 0.49846
256 40.00000 181,159 181,159 0 0.00000
256 50.00000 226,449 226,448 1 0.00044
256 50.00000 226,449 226,449 0 0.00000
256 60.00000 271,739 271,170 569 0.20939
256 60.00000 271,739 271,463 276 0.10157
256 70.00000 317,028 317,028 0 0.00000
256 70.00000 317,028 317,028 0 0.00000
256 80.00000 362,318 362,081 237 0.06541
256 80.00000 362,318 362,081 237 0.06541
256 90.00000 407,608 406,864 744 0.18253
256 90.00000 407,608 138,361 269,247 66.05538
256 100.00000 452,898 451,454 1,444 0.31884
256 100.00000 452,898 0 452,898 100.00000
* 2nd problem, if I start a second test after this one : I get following
results :
so all packets of the 2nd stream are dropped ! while no packets of the
first one are dropped.
It seems as something isn't reset enough, because normally everything
(the queues are empty in between) should be zeroed. I had a little look
in the code but it was too difficult to see the solution at once, so
maybe the author of the red elements knows this immediately. The version
of the element is cvs 04th July but there are no changes in cvs.
256 10.00000 45,289 45,289 0 0.00000
256 10.00000 45,289 0 45,289 100.00000
256 20.00000 90,579 90,579 0 0.00000
256 20.00000 90,579 0 90,579 100.00000
256 30.00000 135,869 135,869 0 0.00000
256 30.00000 135,869 0 135,869 100.00000
256 40.00000 181,159 181,159 0 0.00000
256 40.00000 181,159 0 181,159 100.00000
256 50.00000 226,449 226,449 0 0.00000
256 50.00000 226,449 0 226,449 100.00000
* are there good values for the RED parameters and the queue lengths ?
thanks for any help
best regards,
Brecht
More information about the click
mailing list