We have a 1840 router with 3 FastEthernet interfaces connecting T1, DSL, and other DSL lines. The T1 is the default route. It's getting killed, whenever one or two users start downloading huge files.
We have also configured route maps to route some users to the DSL lines. to offload the T1. But this does not help with occasional large download.
Here is my delimma. Should we configure the FasterEthernet interface connecting to the T1 router to have some sort of intelligent queuing so no one single user can take all the bandwidth? Or should I load-balance among the 3 circuits (how)?
QoS is used to prioritize one type of traffic over another. Without QoS, all traffic streams will share the bandwidth, all of them dropping some traffic. All of the streams will "equalize" based upon the total delay and bandwidth available end-to-end for each stream. TCP uses a "sliding window", or the amount of data that can be in transit without an acknowledgement. Each time a packet is dropped, TCP will cut the TCP window in half. This will reduce the amount of bandwidth used by that stream. For each acknowledgement of a TCP segment with available window, the sender will send an additional MTU worth of data until the window is full. This results in the sharp drop-offs in bandwidth on streams when traffic is dropped, and a slow increase in bandwidth when bandwidth is available. The end result is that the lowest bandwidth streams will suffer the lowest packet drops, and highest bandwidth stream will suffer the most. This mechanism is very democratic, the lowest bandwidth streams will usually get all the bandwidth they need, while the highest bandwidth streams will get the rest. If only a few users are using lots of bandwidth, then you don't have a problem. By rate-limiting users to max amount of bandwidth, users will not see any significant improvement in performance and your users with big downloads will use less bandwidth for longer periods. The only real benefit will be that your T1 wont be running at full bandwidth. QoS would apply if you wanted one group of users, or type of traffic to receive priority over other users or types of traffic. For example, you had a group that if they had a big downloads, that they would receive the lions share of bandwidth over another group with big downloads. We use QoS on our network and have four groups of traffic, priority, high, normal and scavenger. The priority queue is for voice traffic, business applications are classified high, virus-scan and other low priority transfers are in the scavenger queue (lowest) and everything else falls into the normal queue. The queues are setup this way so that during times of congestion, voice traffic is not affected, the business applications get priority over internet and scavenger traffic.
In any event, QoS is not the solution to lack of bandwidth. QoS only prioritizes traffic, and traffic from all queues is affected during times of congestion (except for the priority queue, but that is a special queue that has reserved bandwidth). The lowest priority queues are affected most.
Load balancing traffic over multiple connections is easy if they are all have the same bandwidth, you simple put all of them in your routing table. Load-balancing traffic over links of unequal bandwidth is very difficult and load balancing inbound internet traffic is extremely difficult.
"AXJ" wrote in message news:VY6dnTSRNNvbnBjVnZ2dnUVZ firstname.lastname@example.org...