Discovering Dependencies for Network Management

• network management 9
• dependency 2

http://research.microsoft.com/en-us/groups/nrg/hotnets06.pdf

Leslie Graph, a simple yet powerful abstraction describing the complex dependencies between network, host and application components in modern networked systems.

Lamport ¹: “A distributed system is one in which the failure of a computer you didn’t even know existed can render your own computer unusable”

the absence of tools to identify the components that “can render your own computer unusable”: the implicit web of dependencies among these components exists only in the minds of the human experts running them. The complexity of these dependencies quickly adds up, requiring more help than traditional IT management software provides.

Leslie Graph

the graph representing the dependencies between the system components, with subgraphs representing the dependencies pertaining to a particular application or activity. Nodes represent the computers, routers and services on which user activities rely, and directed edges capture their inter-dependencies.

existing approaches

the system could evolve faster than the rules; deployment of various forms of middlebox (e.g., firewalls, proxies) can change the application’s dependencies without the rule writers even being aware;

expose dependencies by requiring all applications to run on a middleware platform instrumented to track dependencies at run-time [1, 4, 7]. …

heterogeneity defeats most such efforts in practice

While a single vendor might instrument their software, it is unlikely that all vendors will do so in a common fashion;

2 Leslie Graphs and Their Uses

Studies show that ~70% of enterprise IT budgets are spent on maintenance.

Leslie … enabling the following techniques for management and troubleshooting

Fault localization

Reconfiguration planning

Help desk optimization

Anomaly detection

3 implementation

approximating the Leslie Graph using low-level packet correlations.

• Constellation system uses a distributed approach, reactively constructing the Leslie Graph of any node on-demand.
• AND system, which stands for Analysis of Network Dependencies, proactively maintains the approximate Leslie Graph at a centralized inference engine.

3.1 Constellation

local traffic correlations are inferred by passively monitoring packets and applying machine learning techniques.

reference

[1] A.Brown, G.Kar, and A.Keller. An active approach to characterizing dynamic dependencies for problem determination in a distributed environment. In IFIP/IEEE IM, May 2001.

[3] W. Aiello, C. Kalmanek, P. McDaniel, S. Sen, O. Spatscheck, and J. V. der Merwe. Analysis of communities of interest in data networks. In PAM’05, Mar. 2005.

[7] M. Y. Chen, A. Accardi, E. Kıcıman, J. Lloyd, D. Patterson, A. Fox, and E. Brewer. Path-based failure and evolution management. In NSDI’04, pages 309–322, Mar. 2004. TBC