Collaborative Research:EAGER:Large-Scale Distributed Scientific Experiments on Shared Substrate

点击翻译按钮获取中文摘要

The next generation of distributed systems is slated for a huge ramp-up in terms of availability of distributed compute resources and thus shall largely depend for their success on the next generation Internet to be able to offer enough diversified network support for these extremely distributed contexts. Scientific applications, on consolidated resources ranging from private clouds within universities and research organizations, science clouds consisting of various grid resources such as supercomputing centers etc, and commercial clouds such as Amazon, GoGrid, Rackspace, etc, represent a highly distributed application context that shall hugely benefit from considerable diversification of the underlying network substrate.The next generation Internet should allow multiple packet switching network contexts to co-exist over a shared substrate. Each context is allowed to independently optimize its specific context without adversely affecting the performance of other co-located network contexts. To that end, the principal investigators (PIs) present ?elastic pipes,? a dynamic, on-demand, end-to-end, infrastructure-provisioning layer for the Internet. Elastic pipes borrow themselves from elastic clouds, an on demand, dynamic, compute resource leasing architecture popularized by various cloud computing platforms. However, the primary motivation behind the elastic pipe concept is to add architectural support for the realization of multiple isolated ?requirement specific? networking contexts to co-exist over the multi-domain shared infrastructure of the current Internet. ?Isolation? as it applies to the proposed design is different from that in ?virtualization? mechanisms based on two fundamental design constraints that the PIs have imposed:C1. The fundamental packet processing discipline in the native infrastructure has to be preserved.C2. Independent routing domains (commonly known as Autonomous Systems (AS) need to be allowed to control and optimize their individual networks.This project is a proof-of-concept implementation of a new object oriented Internet architecture. This implementation is expected to result in a thorough feasibility study of the basic principles behind the PIs? architectural ideas for the next generation Internet. Additionally, the elastic pipe abstraction is in itself an independent module that can be incrementally implemented over the present Internet to provide huge gains in terms of diversification of the presently ?ossified? Internet.The PIs use the specific use-case context of large scale scientific experimentation to demonstrate the claims of our architecture. Large scale distributed scientific experiments consist of geographically spread out scientists, scientific installations using dedicated networking resources. Such specialized facilities incur huge setup costs and time that need to be amortized through sufficient long term usage. Large scale scientific experiments thus serve as the perfect use case for multiple ?requirement specific networking contexts.?Broader ImpactThe proposed research may serve as a basis of a next generation Internet architecture that will allow natural sharing of resources among multiple organizations by dynamically configuring and creating a requirement specific network context for a particular application. Work under this award will develop the primitives of the ?elastic pipes? abstraction that has the potential to change the basic paradigm of all distributed computing. Also, the basic constraints placed on the design allow an incremental deployment of the PIs? ideas to the present Internet. The PIs plan to organize special sessions at leading business and networking conferences, such as Globecom, Infocom, and SIGCOMM and the results of this project will be integrated in graduate courses on networking architecture at the two universities.