QOS (CBWFQ and Priority)on an 1800 not working

[keiza.hes1]

My test bed includes the following; I have a 1800 router connecting two hosts on which I am trying to use CBWFQ. On one host I have two iperf servers on prts -8800 and -5001. On the other host I have three TCP stream connections; 2 to 8800 and one to 5001 (using Iperf).

I expect to see that my two 8800 streams share (equal bandwidth) a minimum of 15% of the available bandwidth and the 5001 to use a minimum of 60% of the available bandiwith. When I set all three streams in action, I see that all share ~33% of the availble bandwidth violating the "5001 expected behaviour" of a minimum of 60%.

class-map match-all allow-man

match access-group 101

class-map match-all allow-default

match access-group 102

policy-map pm1

class allow-man

bandwidth percent 15

class allow-default

bandwidth percent 60

class class-default

fair-queue

access-list 101 permit tcp any any eq 8800

access-list 102 permit tcp any any eq 5001

interface FastEthernet0/1

ip address 172.22.25.111 255.255.255.0

service-policy output pm1

What am I doing wrong??? Incidentally I cannot get Priority based queuing working either...

[keiza.hes1]

Hi,

The above offered rates suggests the observed bandwidths of ~33% +_~7%. The fact that these offered rates do not reflect my specified bandwidth is a problem. Where do these values originate from???

Iperf servers can (and do) display the throughput received at its end (quite different than the 100Mbps sent). This is how I get the ~33% of the traffic streams. Incidently the bandwidth seen at the server end totals ~6-8MBps not 10Mbps!!! Yet another perculiarity, that I dont understand.

In answer to your question, three machines are sending at 100 mbps (ethernet speed and UDP bandwidth in IPERF) into the router. I presume the router is dropping these to a rate of 100mbps of course.

However there is still 100Mbps into the router and 10Mbps out of the router.

[Joseph W. Doherty]

Offered rates aren't related to specified bandwidth (I believe). They measure bandwidth being sent to the policy map, not the bandwidth being forwarded.

You note you're not seeing 10 Mbps received. but what does the stats for the outbound interface show?

Sending 300 Mbps to a 100 Mbps interface should mean the device upstream of the 1800 is dropping 200 Mbps, but which 200?

As I noted in my prior post, if you're pegging the CPU, feature functioning might not be expected. Depending on the model of 1800, even 3 10 Mbps streams (as I suggested) might overtax the router.