Policy-Driven Autonomic Management in Enterprise-Scale ... - CiteSeerX

14 downloads 1839 Views 42KB Size Report
ing the IFLOW [2] middleware for handling enterprise-scale in- formation flows. The main focus ... is equipped with runtime monitoring capabilities for network de-.
Policy-Driven Autonomic Management in Enterprise-Scale Information Flows Vibhore Kumar∗, Brian F. Cooper† , Greg Eisenhauer, Srihari Govindharaj, Chaitanya Karlekar Mohamed Mansour, Karsten Schwan, Sangeetha Seshadri, Balasubramaniam Seshasayee College of Computing Yahoo! Research† Georgia Institute of Technology 2821 Mission College Blvd. Atlanta, GA 30332 Santa Clara, CA 95054 {vibhore, eisen, srihari, karlekar, mansour, schwan, sangeeta, bala}@cc.gatech.edu [email protected]† 1. Introduction For the past three years we have been involved in developing the IFLOW [2] middleware for handling enterprise-scale information flows. The main focus of our work has been on designing and implementing self-management capabilities for the IFLOW middleware. In the course of its development IFLOW has gradually morphed from a prototype system used for evaluating self-configuration and self-optimization algorithms [3], to an information-flow middleware that supported the concept of utility [4], then to a middleware that incorporated self-healing [1] and now, in its current avatar, the system is capable of using policies for enabling self-management. The IFLOW system at present consists of around 25,000 lines of C++/C code and a Java based GUI has been recently added to the implementation. Figure 1: Screen-shot showing a deployed flow-graph

2. System Capabilities & Features The IFLOW middleware will be running on 16 or more physical nodes located on our instance of Netlab at Georgia Tech. The network underlying these nodes will be representative of one found in enterprises, with presence across the globe, in terms of inter-node delays and bandwidths. We will make use of datastreams from scientific simulations and data-streams that closely emulate the operational information system used by one of our industry partners to drive the demonstration. The Java based GUI will connect to our experiment instance at Netlab and will act as our front-end to the IFLOW nodes. The IFLOW middleware is equipped with runtime monitoring capabilities for network delays, bandwidth and node status, which are then used to drive our self-management features. In particular, we will be demonstrating the following middleware features. Self-Configuration - A user can construct a flow-graph using the drag-n-drop Java GUI, embed code (that will be dynamically compiled at runtime) into the flow path to process/aggregate the events, declare a deployment policy and press the ‘deploy’ button. The middleware will autonomically configure the flow graph across the network of nodes in accordance with the utility policy. The user can then interact with the deployed flow-graph using other GUI features (see Figure 1). Self-Optimization - If the processing load on one of the nodes being used by the flow-graph increases; the flow-graph in response could be reconfigured to use some alternate node. Reconfiguration can be triggered in response to a variety of other events like increased network delay, congestion, etc. The reconfigured flow-graph can be visualized using the GUI and the console prints alert messages corresponding to such actions.

Self-Healing - IFLOW supports check-pointing based recovery from failures. In case of a node failure IFLOW has the capability to recover by re-instantiating the components on another node. This failure recovery can be both stateless or stateful, as is dictated by the flow-graph policies. The self-healing process can be easily visualized using the GUI. Policies - The IFLOW middleware provides the capability to add, delete or modify policies at runtime. These policies can be global to the middleware or specific to the flow-graph. We will be demonstrating these capabilities and show the effect of policies on deployed flow-graphs.

3. Infrastructure Requirements We will need high-speed (broadband or DSL) network connectivity, a location suitable for projection, power supply, easels and tables for the demonstration. It would be very helpful if the organizers can arrange for a projector that connects to a laptop.

References [1] Z. Cai, V. Kumar, B. F. Cooper, G. Eisenhauer, K. Schwan, and R. E. Strom. Utility-driven management of availability in enterprise-scale information flows. Middleware Conference, 2006. [2] V. Kumar, Z. Cai, B. F. Cooper, G. Eisenhauer, K. Schwan, M. Mansour, B. Seshasayee, and P. Widener. Implementing diverse messaging models with self-managing properties using iflow. ICAC, 2006. [3] V. Kumar, B. F. Cooper, Z. Cai, G. Eisenhauer, and K. Schwan. Resource-aware distributed stream management using dynamic overlays. ICDCS, 2005. [4] V. Kumar, B. F. Cooper, and K. Schwan. Distributed stream management using utility-driven self-adaptive middleware. ICAC, 2005.

1

Suggest Documents