Multicast delivery of web pages or how to make web servers pushy

1 Making the Case The Web has gained tremendously in popularity over the last several years. Web administrators are struggling to upgrade their servers to handle the huge volumes of requests. One approach to handling the large volume of requests has been to simply buy more hardware. Other solutions have been investigated including improvements in the HTTP protocol , the use of transparent server replicationn1] and caching of hot pagess2]. At issue is how Web pages can be delivered to increasing numbers of users given limited server capacity and network bandwidth. Another technique that can be used in conjunction with caching and replication to improve scalabil-ity is multicast delivery. With this approach Web pages are delivered to multiple awaiting clients using one server response instance and using underlying network support for point-to-multipoint communication. Although multicast delivery has long been viewed as an approach to provide scalable services 3, 4, 5, 6, 7], its use for multicast Web delivery on an end-to-end basis (multicast from server directly to clients) has received less attention. This has been for mainly three reasons: 1. The Web is largely viewed as a \pull" service where clients' needs for pages are satissed through individualized point-to-point transmission. 2. There is lingering doubt over the usefulness of multicast Web page delivery. The beneets of the server and network aggregation provided by multicast delivery are clear. What is in doubt is whether these beneets may be more than oo-set by the overheads within the server and the network to support multicast delivery. 3. End-to-end multicast is not widely deployed over the commercial Internet. Our work addresses the rst two points above explicitly. First, access to Web sites typically follows a skewed pattern with a small number of hot pages being accessed very frequently, a larger number (but still small) of warm pages being accessed with moderate frequency and a large number of cold pages being accessed infrequently or in some cases rarely. Multiple requests received for the hot and warm pages can be aggregated and responded to once by the server. Multicast delivery support within the network can insure a similar form of bandwidth ag-gregation. An important aspect of our approach is that the system can be built such that clients of a multicast delivery Web server can maintain the illusion that they are interacting with the server according to a pull paradigm. In addition to deening an architecture …