I have a QOS setup that is based on the DSCP Wiki article.

http://wiki.mikrotik.com/wiki/DSCP_based_QoS_with_HTB

I have a couple routers that get bad packet loss as soon as I enable the config.

The config is similar to the wiki and worked great on low usage sites...

It starts with 6-10 rules that filter inbound packets to decide where they are coming from and whether they need to have their TOS bit marked.

After those rules i have two chains. A TOS Bit marking chain of ~80 rules and a packet marking chain with 64 rules. If a packet already has its TOS marked(marked on a previous hop of our network) it only goes though the second chain. If the packet is not marked yet it will go though both chains before hitting the queue tree.

I'm running 1100AHx2 v6.4 Router boards and am experiencing major packet loss on a couple routers and minimal loss on less loaded routers.

I have two 1100AHx2's that run at 40-50% cpu usage and experience 8-10% packet loss. On other 1100AHx2's I experience 1% loss at ~20% cpu usage and 0% loss at < 5% cpu usage.

Interface queues are set to hardware only.
Connection tracking is on.
minimal forward chain filter (blocks bogons and net-bios).
all queue's in queue tree are pfifo.
no bridges or bridge filters.
no nat.
~200Mbs of throughput on the two routers with high packet loss.

QOS config enabled... 40-60% cpu usage...
--- rbA.mycompany.com ping statistics ---
100 packets transmitted, 92 received, 8% packet loss, time 99185ms
rtt min/avg/max/mdev = 1.938/14.593/163.168/24.064 ms

QOS config disabled... consistent 20% cpu usage
--- rbA.mycompany.com ping statistics ---
100 packets transmitted, 100 received, 0% packet loss, time 99145ms
rtt min/avg/max/mdev = 2.025/9.647/177.870/19.018 ms

Some other forum users said it could be a bug in version 6?... Or set hardware only interface queue but they are already set as such.
http://forum.mikrotik.com/viewtopic.php ... queue+type

I get 0% loss when QOS is disabled.

I would like to stay with version 6 if possible.

Please let me know if any other information is needed.

Do I need to revert back to v5 or am I just overloading the router boards or?

which ports handle the load?

If you are using ports 12 or 13 I would recommend switching to other ports, since those are connected thru a PCI-e chip that has low cpu priority.

http://i.mt.lv/routerboard/files/Block-RB1100AHx2.pdf

Ports 1-8 are in use. They are all routed with no bridges or bridge filters. The interfaces are all set to hardware only queues with a queue tree on the outgoing traffic of its wan port only.

Hello,

check /tool profile and /system resources cpu, to see more info about load.

In case you use later-7 filter massively you need to ensure that no unnecessary
traffic goes trough it.

layer-7 works with packet streams that it gets from first several packets - so it
adds latency. One layer-7 filter work under control of one CPU.

In other case you need to find much more powerful device to apply this config on.

Regards,
Janis Megis

I was using "/system resources cpu" to watch the resources and did not realize that "/tool profile" showed exactly what processes are taking the load.

After checking it out, Magis is right. A significant amount of CPU is going toward the four layer 7 rules I was using for filtering p2p and voip traffic.

Once again I'm impressed with the tools that mikrotik gives us to troubleshoot with !

One layer-7 filter work under control of one CPU.

Four layer 7 rules on a duel core router seems to be the biggest issue. With only one layer 7 rule enabled the rule used 2-4%cpu. With all four enabled layer 7 was using 30%cpu just for layer 7 matching. I took the four layer 7 rules out and replaced them with some other rules that are not catching as much traffic but should be sufficient. CPU usage has gone way down. I will report back on the packet loss later as load increases.