Page 1 of 1

M3P doesn't work with bandwidth test

Posted: Mon Oct 17, 2005 2:42 pm
by phendry
I have been testing M3P to see how well the packet aggregation improves things however I can't see it improving anything so far. The tests are using the built-in bandwidth test over an EoIP tunnel. Setup I have is Mikrotik -> Cisco Switch -> Mikrotik and both Mikrotiks can see each other as MNDP neighbors. On the test I set the direction=transmit and local-tx-size=100. I test with packing and unpacking set to simple with an aggregate of 1500 on both ends but get the same results as when I have packing and unpacking set to none.

Is there something I am missing here? Does M3P still work when you run bandwidth test or does it only affect traffic passing through the Mikrotik?

Posted: Mon Oct 17, 2005 4:04 pm
by phendry
Interestingly enough I just roled back to 2.8.28 and it works fine. Problem is that we need "interface bonding" which is only in 2.9. Is this a known bug? Any ideas on a likely fix date?

Posted: Mon Oct 17, 2005 4:22 pm
by normis
yes yes, we are also testing, hold on.

Posted: Mon Oct 17, 2005 5:03 pm
by phendry
yes yes, we are also testing, hold on.
Thanks guys. 8)

Posted: Tue Oct 18, 2005 12:54 pm
by uldis
looks like we have reproduced this problem - we will check what we can do.

Posted: Sun Oct 23, 2005 12:17 pm
by phendry
Any news on this bug?

Posted: Mon Oct 24, 2005 2:17 pm
by uldis
will be fixed in the next release.

Posted: Wed Nov 02, 2005 6:58 pm
by phendry
Good to see a fix for M3P in the latest release. I will try and test this some time this week. Also in the latest (2.9.7) is "fixed queue time limits". Does this refere to the 15ms delay that M3P adds? If so can someone explain how this has changed?

Posted: Thu Nov 03, 2005 10:26 am
by Eugene
No, 15ms is the property of M3P (i.e. it works this way). The fix apparently was for time matcher in queues.

Posted: Thu Nov 03, 2005 6:04 pm
by phendry
Is there no way that the 15ms delay can be user defined or reduced? I know it's the way that M3P does it's thing but it would make sense for the delay to be less. I.e. Under heavy load M3P would easily be able to fill a 1500 byte packet in 5-10ms where as under lite load M3P isn't so important as your not loading the link anyway.