No abstract

Abstract-In the current Internet, many applications start sessions with multiple connections to multiple servers in order to expedite the reception of data. One may argue that such aggressive behavior leads to unfair sharing of bandwidth using the current per-connection rate allocation methods. Sessions with more connections get a higher total rate than competing sessions with fewer connections. In this paper, we explore the issue of fairness of rate allocation from a session point of view. We define a multipoint-to-point session as a set of point-to-point connections started from multiple servers to a client in order to transfer an application-level object. We present session fairness definitions, propose algorithms to achieve these definitions, and compare the resulting allocations with the traditional connection fair algorithm. It is clear from our evaluations that the session fair algorithms proposed achieve a more fair distribution of session rates than the connection fair algorithm, by redistributing the rates claimed by sessions with more connections. We present some initial thoughts on the challenges involved in implementing the session fair algorithms proposed.

Abstract.We consider an overlay network where traffic on an overlay hop is carried in one or more TCP connections established between the overlay nodes at the ends of the hop. We are interested in maximizing the throughput of data carried by this type of overlay network. To that end, we focus on a single path in such a network and investigate how one can determine the number of TCP connections on each overlay hop so as to maximize the throughput of the data carried end-to-end on that path. We first show that having more than one TCP connection on some overlay hops can indeed increase the throughput on overlay paths. We then propose the Adaptive Overlay-TCP Provisioning approach, that, based on the path state, dynamically introduces and removes TCP connections on individual overlay hops to maximize throughput while minimizing the overhead of extraneous connections. We consider two schemes to assess the path state, the intermediate buffer occupancy scheme and the isolated rate probing scheme, and evaluate them experimentally on a set of Planetlab nodes. We show that these schemes can significantly improve the end-to-end throughput with very little overhead.

Content delivery networks (CDNs) currently deliver mostly static and streaming content. However, proxy caches can improve the delivery of these content types as well. A unique value of CDNs could be in improving access to dynamic content, which cannot be cached by proxies. We refer to such a CDN as an Applications CDN, or ACDN. An ACDN will allow a content provider not to worry about the amount of resources provisioned for its application. Instead, it can deploy the application on a single computer anywhere in the network, and then the ACDN will replicate and migrate the application as needed by the observed demand. This demo shows a functional prototype of an ACDN.An ACDN has a fundamental difference with a traditional CDN oriented towards static content. A traditional CDN server is willing to satisfy any request for any content from the subscriber Web site, either from its cache or by obtaining the response from the origin server. In contrast, an ACDN server must have the requested application, including executables, underlying data, and the computing environment, to be able to process a request. Deploying an application at the time of the request is impractical; thus the ACDN can distribute requests only among the servers that currently have a replica of the application; at the same time, the applications must be placed on ACDN servers asynchronously with requests. Thus, ACDN must provide solutions for the following problems that traditional CDNs do not face:Application distribution framework: An ACDN needs a mechanism to deploy an application replica dynamically, and to keep the replica consistent. The latter issue is complicated by the fact that an application typically contains multiple components whose versions must be mutually consistent for the application to function properly.

No abstract