As bandwidth has increased, the Internet has become a large hub for streaming media between users. The media types being delivered include Voice Over IP communication, live TV shows and web based radio among many others. In some cases, a user participating in a media session might want to record the communication taking place. This is usually implemented by a client side recording application which receives the same data as the client and writes it to disk. This allows for any number of concurrently recording users since each client is in charge of their own recording. Each clients machine will be utilized for both processing and storage which will scale up to any number of users, as long as they have enough room for storage. Client-side media recording does however have a number of disadvantages which can be avoided by having a central server take care of the recording instead. Examples of these include the inability to easily apply updates to the recording functionality as well as requiring the clients to have sufficient storage space. This Masters Thesis examines the advantages and disadvantages in terms of scalability and usability of having a server-side recording facility instead of distributing the load onto client machines. It also covers playback of the recorded streams, and examines the differences with playing back server- side media, as opposed to media recorded at a clients machine. Examination shows that client-side recording is easier to scale for large groups of users, but server-side recording is a viable option. Server-side recording requires more engineering and resources, but does also provide better control over the recording operations compared to when they are distributed onto clients. To provide proof of concept of a scalable recording facility, a prototype was developed which supports recording, listing and playing back media streams. The implementation has so far only been tested at a smaller scale, but is built to scale linearly with the amount of concurrent users and is expected to work at a larger scale as well.