Scenario: ISP adding streaming media caches

About this scenario

The ISP in this scenario provides Web caching service to its customers. The ISP has Web caches in its POPs and in the Data Center. The ISP now wants to add streaming media service.

Currently, all traffic from the POPs is routed through the Data Center. The ISP wants to continue with this routing scheme when streaming media service is added.

ISP's requirements

The ISP's requirements when deploying streaming media service are as follows:

To limit bandwidth consumption by streaming media
The deployment of Web caches in the POPs significantly reduced Web traffic across the WANs to the Data Center. The ISP wants to implement a similar scheme for streaming media. The reason is that the ISP expects that the addition of streaming media service will cause traffic congestion across the WANs, resulting in slow service for all protocols. Additionally, the ISP managers want a deployment scheme that does not mandate purchasing additional bandwidth.
To provide good quality streaming media service for customers using modems of various speeds
The ISP's customers have modems of various speeds, with many customers still using 28.8-kbps modems. The ISP wants to be able to provide the best-quality streaming media service that is possible for each of the common modem speeds.

Deployment illustrated

The following illustration shows how the ISP deployed streaming media caches. For simplicity, only two POPs are shown in the illustration.

Note
For this scenario, all traffic from the POPs to the Internet is routed through the Data Center.

Deployment at the Data Center:

A hierarchical deployment was created by adding streaming media caches in the Data Center to handle streaming misses from the POPs. This deployment significantly reduces the number of connections to the streaming server. The reason is that if multiple POPs request the same stream, a Data Center streaming media cache needs to connect to the streaming server only once for each unique streaming media. For live streaming media, the cache can then respond to additional requests for the same unique stream by splitting the stream for any additional requests for the same stream.
All NetCache appliances at the Data Center are dedicated streaming media caches or dedicated Web caches.
In this scenario, using dedicated appliances in the Data Center is especially desirable because the Data Center NetCache appliances must handle high volumes of each type of traffic from many POPs.
Multiple streaming media caches were deployed at the Data Center for redundancy. The ISP managers do not want streaming service to be unavailable if a streaming media cache goes down.
The managers realize that if a streaming media cache goes down, any connections between a streaming server and the cache that are open will be lost. However, subsequent connection attempts could be handled by another streaming media cache.

The number of streaming media caches deployed in the Data Center was determined by considering the following:

The volume of streaming traffic that the managers expect at the Data Center
No clients send streaming requests to the Data Center streaming media caches directly. All streaming traffic is a result of streaming misses at the POPs.
The need for redundancy
The capability of the network connection to the Internet to handle streaming requests
The more streaming media caches deployed, the more separate connections to streaming servers in the Internet would be established.

Because the ISP does not own any streaming servers, shielding the streaming server from too many connections is not an issue for the ISP. A company that owns the streaming server and is setting up streaming service would need to consider the number of connections that the streaming server could handle.

Deployment at the POPs:

The Web caches that were previously deployed at the POPs remain dedicated to Web caching.
One dedicated streaming media cache was deployed at each POP to reduce the amount of streaming traffic over the WANs.
The L4 switch at each POP sends streaming requests to the Data Center if the streaming media cache at the POP goes down. In this case, the L4 switch at the Data Center passes the streaming requests to a streaming media cache at the Data Center.
A second streaming media cache could have been deployed at each POP to provide failover locally in case the first streaming media cache went down. If all streaming media caches go down, the L4 switch at the POP sends the streaming requests to the Data Center.

As an alternative, configuring a hierarchy on the two streaming media caches in the POP could eliminate the need for an L4 switch in the data center. If a hierarchy was configured on a POP streaming media cache, cache misses would be sent from the POP cache up the hierarchy to the logical parent cluster in the data center. The drawback is that if both streaming media caches in the POP became unavailable, client traffic would be sent directly to the Internet instead of to a streaming media cache in the data center.