[SOLVED] traffic shaper queues bug

xbipin

to narrow it down further, i think its got to do with selecting UDP as protocol coz i created new rules under floating tab and assigned same queue and those work fine for some reason, image below

CropperCapture[1].jpg
CropperCapture[1].jpg_thumb

bio_hazard

Is definitely something with UDP traffic, i opened a ticket on the bug tracker. http://redmine.pfsense.org/issues/1546

Lets wait for a answer.

xbipin

this issue still remains, i cleared all rules and recreated them along with queues based on the reply i got
http://redmine.pfsense.org/issues/1582

xbipin

any1 willing to have a look at my config file in order to help me sort out this issue if it is really config related coz my voip phone seems to be almost unusable at high traffic and might have to switch to a old snap to make the queues work properly?

danswartz

You seem to be having the exact behavior I see. Unfortunate the ticket just got closed, but this is supposedly be addressed by a pre-GA change (or so I seem to recall reading…)

dusan

Some peoples prefer to call it a feature or config issue, not a bug. In my view this should be termed a design bug/error as this violates the principle of stateful traffic processing.

In a stateful firewall/router/shaper users must be able to filter/route/shape traffic simply by specifying the behavior with respect to the session-initiating packet. Take for example NTP, the administrator must be able to pass, to route and to shape incoming NTP requests to the correct NTP server and the outgoing NTP replies from that server as well without special care of the 'outgoing' direction.

xbipin

can u plz elaborate whats wrong as having a UDP queue which as a matter of fact used to work perfect earlier but something broke it, if i only knew which commit was it then would manually revert those changes on my box.

dusan

I can't tell. I just guess that it cannot be undone simply by reverting a single patch.

xbipin

wouldnt it be possible to edit the config file manually to change the behavior temporarily till its fixed?

dusan

I have no clue on manual config file editing. A temporary workaround (that appears to work for me) is to simply ignore the Floating tab and to select queues directly on LAN, WAN and all OPTx tabs. It is not comfortable, but doable.

xbipin

i have the below rule which i tried match as well as pass but still download traffic doesnt go to the proper queue

	 <rule><id><type>pass</type>
		 <tag><tagged><direction>out</direction>
		<quick>yes</quick>
		<floating>yes</floating>
		 <max><max-src-nodes><max-src-conn><max-src-states><statetimeout><statetype>keep state</statetype>
		 <os><protocol>udp</protocol>
		<source>
			 <any><destination><address>mysip</address></destination> 

		<defaultqueue>qVoIP</defaultqueue></any></os></statetimeout></max-src-states></max-src-conn></max-src-nodes></max></tagged></tag></id></rule> 
	 <rule><id><type>pass</type>
		 <tag><tagged><direction>in</direction>
		<quick>yes</quick>
		<floating>yes</floating>
		 <max><max-src-nodes><max-src-conn><max-src-states><statetimeout><statetype>keep state</statetype>
		 <os><protocol>udp</protocol>
		<source>

<address>mysip</address>

		 <destination><any></any></destination> 

		<defaultqueue>qVoIP</defaultqueue></os></statetimeout></max-src-states></max-src-conn></max-src-nodes></max></tagged></tag></id></rule>

can u plz mention this above rules how do i recreate using wan and lan tabs to make it work

dusan

Following is my config for a teleconference device that use TCP for control channels and UDP for data channels. Not sure if it works for you.

– on the LAN interface add new pass rules before the default rule with source IP = any, destination IP = any, destination port = 1720 TCP and 3230-3279 UDP. Select queue = qVoIP.

– repeat the same procedure on the WAN and OPTx interfaces. (Note: my WAN + OPTx are all connected to Internet. I've started setting up the shaper using the single-LAN-multi-WAN wizard.)

I've configured the shaper this way only for teleconference. In my system there are so many Internet links and so many types of traffics that I must rely on the (now unreliable) Floating Tab for the other traffic types. At least untill we replace pfsense with an other shaper.

xbipin

i think it wont work as im using rules that match based on the server ip rather than any particular or range of ports but ill still give it a try

eri--

I am posting here but this is in general.
Traffic shaper behaviour should be ok on latest snapshots.

bio_hazard

@ermal:

I am posting here but this is in general.
Traffic shaper behaviour should be ok on latest snapshots.

I updated to the latest available (built on Mon Jul 4 16:48:37 EDT 2011) but didint change nothing to me.

danswartz

Thanks, Ermal. I will try to update tonight after my wife goes to bed :) Currently running snap from June 22nd. This is one reason I love having pfsense virtualized on my ESXi server.

1. Take snapshot of virtual machine on the ESXi box.

2. Do the upgrade, and reboot pfsense.

3. Test things out.

4a. If all is well, delete the snapshot.
4b. If something is pooched, roll back to the snapshot and reboot pfsense.

dusan

@Ermal: I updated to Mon Jul 4 16:48:37 EDT 2011 snapshot but still it does not seem to work. I have an NTP server for testing. With the Floating Tab alone using pass out on any interface rules (with queueing), only NTP requests to my server are shaped, NTP replies are not.

danswartz

No detectable change at all. I wiped the shaper. Created the same trivial shaper config as before. e.g. 10.0.0.7 is the voip host. Set for PRIQ. Save the config. Call my office phone and leave a message. No activity on the qVOIP queue. Finally, out of desperation, I deleted the 10.0.0.7, and saw the one floating rule replaced by two rules: one for the SIP ports and one for the RTP ports. I called my office phone again, and started leaving a voicemail. Looking at the queues as I did, I saw about 86kb/sec going out via qVOIP, so it is clearly the source IP address matching that was hosing me. Sigh…

danswartz

Oh yeah, and the failure to use qACK at all outbound is still there…

xbipin

i guess the last snapshot built was 04-Jul-2011 18:11 and in that its not fixed but mayb if a new snap is built then need to test on that