Despite the advance of network technologies in the past decade, providing video services to a large number of users remains a major technical challenge. This is especially true when it comes to serving high-definition videos. This thesis makes two contributions towards providing large-scale and cost-effective video services. 1) We consider the problem of periodic broadcast of popular videos in client/server video systems and present two novel techniques. Our research advances the state of the art with a segmentation rule that can generate a series of broadcast designs, among which we can choose the one that results in the smallest broadcast latency. We show that this rule allows us to design the broadcast technique that is the fastest up to

date. 2) We then look at the problem of service scheduling in fully distributed peer-to-peer video systems, where a large number of hosts collaborate for the purpose of video sharing. Our proposed technique allows a client to be served by a server that is beyond its own file look up scope and can dynamically adjust client and server matches as new video requests arrive in the system. Our performance evaluation shows that these features dramatically improve the system performance to a large extent in terms of reducing service latency under a range of

simulation settings.