There are two basic steps to optimize upstream bandwidth. First we have to find a way to prevent the ADSL modem from queuing packets since we have no control over how it handles the queue. In order to do this we will throttle the amount of data the router sends out eth0 to be slightly less than the total upstream bandwidth of the ADSL modem. This will result in the router having to queue packets that arrive from the Local Network faster than it is allowed to send them.
The second step is to set up priority queuing discipline on the router. We'll investigate a queue that can be configured to give priority to interactive traffic such as telnet and multi-player games.
The final step is to configure the firewall to prioritize packets by using fwmark.
Although the connection between the router and the modem is at 10Mbit/s, the modem is only able to send data at 128kbit/s. Any data sent in excess of that rate will be queued at the modem. Thus, a ping packet sent from the router may go to the modem immediately, but may take a few seconds to actually get sent out to the Internet if the queue in the modem has any packets in it. Unfortunately most ADSL modems provide no mechanism to specify how packets are dequeued or how large the queue is, so our first objective is to move the place where the outbound packets are queued to somewhere where we have more control over the queue.
We'll do this by using the HTB queue to limit the rate at which we send packets to the ADSL modem. Even though our upstream bandwidth may be 128kbit/s we'll have to limit the rate at which we send packets to be slightly below that. If we want to lower the latency we have to be SURE that not a single packet is ever queued at the modem. Through experimentation I have found that limiting the outbound traffic to about 90kbit/s gives me almost 95% of the bandwidth I could achieve without HTB rate control. With HTB enabled at this rate, we've prevented the ADSL modem from queuing packets.
At this point we still haven't realized any change in the performance. We've merely moved the FIFO queue from the ADSL modem to the router. In fact, with Linux configured to a default queue size of 100 packets we've probably made our problem worse at this point! But not for long...
Each neighbor class in an HTB queue can be assigned a priority. By placing different types of traffic in different classes and then assigning these classes different priorities, we can control the order in which packets are dequeued and sent. HTB makes it possible to do this while still avoiding starvation of any one class, since we're able to specify a minimum guaranteed rate for each class. In addition to this, HTB allows for us to tell a class that it may use any unused bandwidth from other classes up to a certain ceiling.
Once we have our classes set up, we set up filters to place traffic in classes. There are several ways to do this, but the method described in this document uses the familiar iptables/ipchains to mark packets with an fwmark. The filters place traffic into the classes of the HTB queue based on their fwmark. This way, we're able to set up matching rules in iptables to send certain types of traffic to certain classes.
The final step in configuring your router to give priority to interactive traffic is to set up the firewall to define how traffic should be classified. This is done by setting the packet's fwmark field.
Without getting into too much detail, here is a simplified description of how outbound packets might be classified into 4 classes with the highest priority class being 0x00:
Mark ALL packets as 0x03. This places all packets, by default, into the lowest priority queue.
Mark ICMP packets as 0x00. We want ping to show the latency for the highest priority packets.
Mark all packets that have a destination port 1024 or less as 0x01. This gives priority to system services such as Telnet and SSH. FTP's control port will also fall into this range however FTP data transfer takes place on high ports and will remain in the 0x03 band.
Mark all packets that have a destination port of 25 (SMTP) as 0x03. If someone sends an email with a large attachment we don't want it to swamp interactive traffic.
Mark all packets that are going to a multi-player game server as 0x02. This will give gamers low latency but will keep them from swamping out the system applications that require low latency.
Mark any "small" packets as 0x02. Outbound ACK packets from inbound downloads should be sent promptly to assure efficient downloads. This is possible using the iptables length module.
Obviously, this can be customized to fit your needs.
There are two more things that you can do to improve your latency. First, you can set the Maximum Transmittable Unit (mtu) to be lower than the default of 1500 bytes. Lowering this number will lower the average time you have to wait to send a priority packet if there is already a full-sized low-priority packet being sent. Lowering this number will also slightly decrease your throughput because each packet contains at least 40 bytes worth of IP and TCP header information.
The other thing you can do to improve latency even on your low-priority traffic is to lower your queue length from the default of 100, which on an ADSL line could take as much as 10 seconds to empty with a 1500 byte mtu.
By using the Intermediate Queuing Device (IMQ), we can run all incoming packets through a queue in the same way that we queue outbound packets. Packet priority is much simpler in this case. Since we can only (attempt to) control inbound TCP traffic, we'll put all non-TCP traffic in the 0x00 class, and all TCP traffic in the 0x01 class. We'll also place "small" TCP packets in the 0x00 class since these are most likely ACK packets for outbound data that has already been sent. We'll set up a standard FIFO queue on the 0x00 class, and we'll set up a Random Early Drop (RED) queue on the 0x01 class. RED is better than a FIFO (tail-drop) queue at controlling TCP because it will drop packets before the queue overflows in an attempt to slow down transfers that look like they're about to get out of control. We'll also rate-limit both classes to some maximum inbound rate which is less than your true inbound speed over the ADSL modem.
We want to limit our inbound traffic to avoid filling up the queue at the ISP, which can sometimes buffer as much as 5 seconds worth of data. The problem is that currently the only way to limit inbound TCP traffic is to drop perfectly good packets. These packets have already taking up some share of bandwidth on the ADSL modem only to be dropped by the Linux box in an effort to slow down future packets. These dropped packets will eventually be retransmitted consuming more bandwidth. When we limit traffic, we are limiting the rate of packets which we will accept into our network. Since the actual inbound data rate is somewhere above this because of the packets we drop, we'll actually have to limit our downstream to much lower than the actual rate of the ADSL modem in order to assure low latency. In practice I had to limit my 1.5mbit/s downstream ADSL to 700kbit/sec in order to keep the latency acceptable with 5 concurrent downloads. The more TCP sessions you have, the more bandwidth you'll waste with dropped packets, and the lower you'll have to set your limit rate.
A much better way to control inbound TCP traffic would be TCP window manipulation, but as of this writing there exists no (free) implementation of it for Linux (that I know of...).