This thread drifted from router discussions:
John,
I think your description of traffic congestion at the ISP level is essentially correct. And I also seriously doubt they honor DSCP; rather they use their rather large BW to deliver traffic at maximum speed up to your paid-for limit. Also your description of TCP vs UDP I think is correct.
In my limited experience, a good ISP will have enough internal BW that their network does not become a QOS issue, hence strict adherence to DSCP tags is generally not required at their level. It is left to the customer to properly manage their use of ISP BW, something that QOS in both directions helps a VoIP/data customer optimally implement.
Here is where I think inbound QOS helps: your router will send congestion packets upstream to your ISP if an incoming TCP connection is threatening to interfere with the prioritized packets. This relieves the user of trying to set a fixed limit of 90% available BW, instead letting the router dynamically use Internet Protocol to control traffic as intended, alloting full BW to all uses of the WAN and LAN as specified in the QOS list.
I have observed the behavior of VoIP calls with simultaneous large BW data useage, and observed the effects of QOS in my Draytek 2130 router. I can hear the effect on phone calls and see the BW priorities dynamically shift. In both directions, of course. When I had a DGL-4300 router I had to do what others have referred to as "redneck QoS", limiting incoming BW useage by all non-VoIP users to a percentage of the total so there was always some incoming BW available for VoIP. A static, rather than dynamic allocation that essentially reduces your BW all the time to assure VoIP performance during a phone call. Trouble is you give up BW all the time to allow for the occasional phone call(s).
Dan, you are correct. The only way you are going to get QOS from a provider is with a SLA.
TCP traffic just flows along the Internet until one of the routers that it hits can't deliver the traffic as quickly as it is coming in. This happens by design at the ISP's last hop before it hits your business. If you have a 10 mbps download speed, they have programmed their last hop to deliver that rate to you. The data you requested may be arriving at that router at 100 mbps. That router then sends congestion packets, a type of ICMP traffic I believe, back to the source to get it to slow down.
So, lets say you set the maximum rate that your router can receive to 9 mbps. Now, when you download pipe gets full, the congestion packets are issued from your router, instead of the ISP's last router, and tells the TCP traffic to slow down. None of this affects UDP traffic. Since UDP traffic is, by definition, best efforts delivery, there is no slowing it down. It is just going to come barreling right on it.
So, if you have throttled your router at about 10% less than the ISP provides, you have left a 1 mbps window for UDP traffic, which you cannot stop anyway.
I have no idea if this is the correct explanation why this works, but where I have implemented this using PFSense as my endpoint, it seems to work.
John,
I think your description of traffic congestion at the ISP level is essentially correct. And I also seriously doubt they honor DSCP; rather they use their rather large BW to deliver traffic at maximum speed up to your paid-for limit. Also your description of TCP vs UDP I think is correct.
In my limited experience, a good ISP will have enough internal BW that their network does not become a QOS issue, hence strict adherence to DSCP tags is generally not required at their level. It is left to the customer to properly manage their use of ISP BW, something that QOS in both directions helps a VoIP/data customer optimally implement.
Here is where I think inbound QOS helps: your router will send congestion packets upstream to your ISP if an incoming TCP connection is threatening to interfere with the prioritized packets. This relieves the user of trying to set a fixed limit of 90% available BW, instead letting the router dynamically use Internet Protocol to control traffic as intended, alloting full BW to all uses of the WAN and LAN as specified in the QOS list.
I have observed the behavior of VoIP calls with simultaneous large BW data useage, and observed the effects of QOS in my Draytek 2130 router. I can hear the effect on phone calls and see the BW priorities dynamically shift. In both directions, of course. When I had a DGL-4300 router I had to do what others have referred to as "redneck QoS", limiting incoming BW useage by all non-VoIP users to a percentage of the total so there was always some incoming BW available for VoIP. A static, rather than dynamic allocation that essentially reduces your BW all the time to assure VoIP performance during a phone call. Trouble is you give up BW all the time to allow for the occasional phone call(s).
Last edited by a moderator: