A systematic approach for migrating enterprise networks
Enterprise network operators must frequently change the design of their networks to reflect new organizational needs. Migrating enterprise networks is challenging as there exists a huge semantic gap between high-level objectives that operators have for their networks and low-level configurations on hundreds of devices, and as there exists complex dependencies among network components and devices. Today the migration is manual and ad-hoc, which has been known to result in overly complex and kludgy designs, and design faults and configuration errors causing outrages and enabling cyber attacks. The challenges in migrating enterprise networks pose a fundamental barrier to the adoption of new and innovative technologies such as cloud computing. ^ The goal of this thesis is to eliminate the barriers that exist today in migrating enterprise networks. We present two fundamentally new and complementary approaches. First, we present a systematic approach to migrating existing networks based on task-driven abstractions. The approach centers around abstracting and formulating each migration task in terms of their network-wide performance, security, and resilience requirements. The formulations capture the correctness and feasibility constraints, and model each task as one of optimizing desired criteria subject to the constraints. Second, we present a top-down framework for modeling and reasoning about network complexity. The framework enables operators to explicitly control complexity when designing new networks as well as migrating existing networks, to minimize the effort and risk involved in future migrations. We devise abstractions to reason about whether and how a combination of design primitives will meet the high-level objectives, and decompose a design into its constituent primitives and quantitatively estimate the complexity of individual primitives. ^ We apply the two approaches to three typical yet distinct enterprise management tasks: (i) evolving and redesigning virtual LANs; (ii) adopting cloud computing to deliver existing enterprise applications; (iii) modeling and controlling the complexity of routing design. Evaluations on operational campus networks show the applicability and effectiveness of our approach. ^
Sanjay G. Rao, Purdue University.