As the distributed computing system becomes widely used today, knowing the internal structure of a network helps people to understand and maintain the network efficiently. Network topology, which extends the topology from mathematics, is used to illustrate the layout of a network, both physically and logically. The devices in network are regarded as nodes, links between them as lines, and they are connected together geometrically to indicate the configuration of the network. The way that they are connected is the topology of that network.
Network topology can be further divided into two types: physical and logical. Physical topology focuses on the physical connection between the devices in a network, while logical topology focuses on the internal data flow regardless of the location of the devices or distances between them1.
A point-to-point topology is simply connecting two nodes directly.
All the nodes are connected to one central line (the bus) in a bus topology. Data is transmitted in both ways along the central line until the node address matches the the destination address contained in the data. If they don’t match, the data is passed to the next nodes without manipulation to ensure that it is received by the right node.
There exists a central hub which all the other nodes are directly connected to and receive datas from. This simple structure also leads to an obvious weakness, once the central hub performs poorly or even breaks down, the rest of the network will be affected.
A ring topology can be regarded as an end-to-end chain, where every nodes is either directly or indirectly connected to the others. Data is transmitted in one direction along the ring, looking for the right destination address.
There is a direct path between every pair of nodes in a fully connected mesh topology, which makes the data transmission faster by choosing the shortest path.
A tree network combines star topology and bus topology together. It has a central cable (the bus) as the backbone, with several branches coming out of both sides, each of the branches is connected to the central hub of a star network2.
There is a combination of at least two different physical topology structures in a hybrid topology. It is commonly used in the network with relatively large scale because of its high cost, on the other side, it combines the advantages of the topologies that are being used to make the whole network run more efficiently and stably3.
One issue that should be considered before building a CDN is how densely the servers should be distributed. The choice of CDN topology – the distribution of CDN servers – will lead to differences in performance. The two main approaches nowadays are the co-location approach and the network-core approach1. The servers of a CDN following the former approach are widely scattered geographically, while the latter are consolidated.
A scattered CDN usually has a large number of surrogate servers distributed pervasively around the target area, with each of them having a relatively low capacity, the proximity to the end-users is therefore maximised2. One of the biggest CDN providers, Akamai, has a highly scattered CDN network. It claims to cover 110 countries with 200,000 servers within over 1,400 networks3.
This page was written by Michelle Zhao.
Amazon launched CloudFront in 2008 as their answer to existing CDN services like Akamai and Limelight1. CloudFront has servers located in Europe, Asia, Australia, South America, and the United States. The map below shows where CloudFront’s edge servers are located.
Notice how the servers are positioned specifically in areas with good connectivity to the internet. ‘Hubs’ include East Asia, Western Europe, and the USA - these are all areas with high internet usage2. In total, Amazon operates 52 edge servers.
CloudFront provide higher performance with algorithms that cache copies of the most popular files. It also ensures that the users are serviced by the closest available edge location with that file3. Another way Amazon manages to keep speeds fast and latency low is by having persistent connections open with their origin servers. This means that if the edge server doesn’t contain the required file, then it can still be fetched relatively quickly from one of the origin servers3.
CloudFront is one of the newer generations of CDNs in that it is optimised for not just static content, but video streaming, file downloading, amongst other things.