Getting Enough Bandwidth with Better Routing
A common question for network connectivity is "how much bandwidth is enough?" While bandwidth can boil down to upload and download limits from an ISP, the overall question is really one of network speed - how fast is the network connection to the Internet? Network speed is greatly influenced by the routing methods used to direct network traffic; the better the traffic routing, the better the network connection's speed and reliability.
Routing Traffic: BGP
The Internet is made up of millions of individual servers, which are all interconnected, like houses on main streets and bystreets. Some central networks, called backbones, have connections to millions of servers, which makes it easier to send traffic along those routes.
If an ISP is connected to a single Internet backbone, then they only have one route to use to send traffic. However, this means if that backbone goes down, the network connection goes down. For redundancy, most ISPs have connections to at least two backbones, and traffic is routed between those backbones.
When there are multiple backbones to choose from, there has to be some way to identify the most efficient route. The most common routing logic is border gateway protocol (BGP) which counts the number of networks (autonomous system, or AS, hops) that each route has between the starting server and the destination server.
For example, one person wants to send an email from their home account with SBC Global to their friend at MSN. SBC is on one network, and MSN is on another. Route A reports six AS hops between SBC and MSN, and Route B has three AS hops between them. With BGP routing, the traffic is sent on Route B.
Performance-Based Routing
The big limitation of BGP is that it only gauges AS hops rather than other more important factors, like latency. Route B from the last example has only three AS hops, meaning traffic only has to cross three networks to reach its destination. However, BGP routing only counts AS hops - it cannot account for the number of individual routers in an autonomous system. Route B may only cross three networks, but if Network 1 has five routers, Network 2 has four routers, and Network 3 has eight routers, the traffic has to go through 17 routers before it can reach its destination. If Route A crosses six networks but each network only has a single router, the traffic only has to pass through six routers. That means that Route A may be significantly faster than Route B, but BGP routing has no way to recognize that, so it sends traffic down the slower route.
Performance-based routing offsets the limits of BGP, and takes a more intelligent approach to traffic routing, by looking at other factors than AS hops:
• Performance metrics like latency, jitter, and packet loss
• Current network load
• Connection type, such as T1 or OC3
Latency can be either the time to send a packet one-way or the round-trip time, like the time to send ICMP packets (ping) to one server and receive the response. Jitter is the fluctuation in latency times. For example, if the first trip time is 3ms and the next is 105ms and the next is 20ms, there is a large swing between trip times, and, therefore, there is high jitter on the connection. Packet loss is the number of packets (information) which never reach their destination. The current load is how much traffic is currently on that connection, and the connection type indicates how much traffic the network can handle effectively.
By looking at the actual quality of the network connection, performance-based routing can select much faster, more reliable routes.
What Better Routing Really Means
Poor routing can corrupt or interrupt packets, requiring information to be resent and increasing the overall time it takes to do anything on the network. While simple tasks like web browsing may not be affected by poor network performance, a number of vital applications can be impaired by poor routing protocols in ways that may not be apparent in a simple upload/download size summary:
• Any kind of media application, such as streaming video or audio
• Potentially business-critical applications like voice over IP (VoIP) or video conferencing
• Upload and download times
• Email delivery
• Remote network applications like VPNs
<< Home