Hierarchy deployment examples

Example 1: cluster used with transparent Web caching

The ISP in this example has two goals:

Implement transparent Web caching so that users do not have to configure their Web browsers to support caching.
If the NetCache appliance at the POP does not have a requested object, a cache miss will be sent to the data center for processing instead of having the NetCache appliance at the POP fetch objects from the Internet directly.

This following illustration shows the ISP's deployment, in which transparent proxying was deployed only at the POPs.

Configuration

The NetCache appliance at each POP was configured to view the NetCache appliances at the data center as a cluster. This enables the NetCache appliance at each POP to distribute requests over the data center NetCache appliances, using the NetCache hashing function. Because the cluster acts as a parent to the NetCache appliance at each POP, the NetCache appliances in the cluster can fetch objects from the Internet, but the NetCache appliances at the POPs cannot.
The NetCache appliances at the data center were not configured to be aware of each other; that is, they are not aware that the NetCache appliance at each POP considers them to be in a cluster, and they do not know that the other NetCache appliances in the cluster exist. The reason is that after a NetCache appliance at the data center receives a request from a POP's NetCache appliance, it is more efficient for a data center appliance to send a request directly to the Internet if it cannot resolve the request rather than try to find the object through intercache communication.

Why transparent proxying was deployed only at the POPs

The ISP's goal of making caching transparent to the user is fulfilled by implementing transparent proxying only at the POPs because all the ISP's customers dial in to one of the POPs.
If a POP NetCache appliance does not have an object, it requests the object from the data center cluster. Even if a NetCache appliance in the cluster must obtain the object from the Internet, the object is then available to any POP that requests it. After the data center's NetCache appliance returns a requested object to a POP's NetCache appliance, the next request a client makes for that object can be resolved by the NetCache appliance at the POP. This process is more efficient and faster than if a POP's NetCache appliance had to send requests directly to the Internet to obtain objects it does not have, and it saves bandwidth.
If a POP has multiple routes, the routes can flap or be asymmetric, which breaks transparent proxying. If multiple routes exist, it is better to use hierarchical caching to avoid these problems. The reason is that it is more difficult to constrain routing at a POP if you have transparent proxying at the data center.
If transparent proxying with an L4 switch was deployed at both the POPs and the data center, noncacheable objects would be processed by two caching sites before the object is requested from the Internet, which is not efficient.
Alternatively, you can use an L7 switch instead of an L4 switch. You can configure an L7 switch so that it obtains noncacheable objects from the Web server directly, bypassing the NetCache appliance completely. However, this approach does not fulfill the goal of preventing direct access to the Internet from the POPs.

Note
You can compare this example to Example 4: POP and data center using HTTP transparent proxying, which shows a switch at the data center and transparent proxying enabled on all the NetCache appliances at the data center.

Where to get more information

You can use hierarchies and clusters to achieve many different results. This section provides a brief introduction to the subject. The chapter about hierarchies in the Administration Guide provides detailed information about using and configuring clusters and hierarchies. It describes ways to customize request distribution to handle various situations, such as forcing certain NetCache appliances to handle specific types of requests and bypassing parent caches for requests to Web servers that are close to a NetCache appliance.