دانلود رایگان ترجمه مقاله مدیریت شبکه سازمانی با KnowNet (نشریه IEEE 2013)

KnowNet : به سمت یک برنامه دانش برای مدیریت شبکه سازمانی

KnowNet: Towards a Knowledge Plane for Enterprise Network Management

Abstract

Network management tasks remain tedious and error-prone, and often require complex reasoning on the part of the network administrator. With KnowNet we address the challenge of reasoning about network management by approaching it as a set of cooperating applications executing over a knowledge graph which captures data and information about the network and the applications that manage and reason over it. We apply our approach to enterprise network management by developing a suite of cooperating applications that deals with security and application performance management in an enterprise network.

1 Introduction

Despite longstanding recognition by the networking research community of the importance of an improved approach to network management and the need for automated solutions [1], the state of the art continues to resist these efforts. In practice network management remains manual, tedious and error-prone, and often requires complex reasoning on the part of the network administrator. Automating the reasoning done by domain experts is the key challenge in realizing a more systematic approach to network management. This is difficult, largely in part because this reasoning must take into account large and diverse quantities of data. This data is collected from a number of different sources, and in order to be meaningfully interpreted, it must be correlated with the operational state of the network at the time it was collected. Complex relationships between hardware, protocols, services, and their operational state must be untangled in order to fully understand faults and effect change in the network.

We argue that a network management system that can support such functionality requires the means to capture network data and knowledge of the network in a holistic manner, allows for network management applications to easily interact with and reason about this knowledge, and to effect change in the network itself. In our work we take a pragmatic approach towards realizing such a system. Our key insight is that network administrators already deploy a multitude of systems to monitor and manage their networks, and that they (the administrators) are the reasoning and logic that tie these systems together. Rather than imposing a new network management system on network administrators that might require them to replace the tools and systems they already use, we propose to provide them with a framework that can turn the systems they are already using into a cooperative whole and allow them to add to that whole by easily automating the reasoning tasks they are currently performing manually.

Towards this end, we present KnowNet, a knowledge-centric network management system that facilitates knowledge sharing and discovery, thus enabling operators to more effectively reason about network state, determine appropriate actions, and ‘write’ to the network to effect needed change. Applications use the knowledge graph at the core of KnowNet to represent data collected about the network and knowledge derived from that data. Knowledge graphs have gained popularity in a number of domains [2], [3], [4], [5], due to their ability to express and discover interesting relationships over large datasets, enabling complex reasoning about interrelated data [2], [6], [7]. In our work on KnowNet, we explore the use of a knowledge graph to capture information and knowledge in a network management setting. As such, our work is a modest step towards a practical (single domain) knowledge plane [1].

The knowledge graph in KnowNet is used to capture more than just data about the network—specifically, network management applications in KnowNet interact with each other and the network itself via the knowledge graph. This allows network management actions to also be captured as part of the holistic network view in KnowNet. To realize such a collaborative network management system, we have developed a knowledge graph, called Kilo1 , in order to explore features that are particularly useful in the context of network management, and may not exist in such a combination in available knowledge graph implementations. Specifically, Kilo includes the typical set of knowledge graph primitives: insert, query, and delete. In addition to these, Kilo also supports a subscribe primitive to learn about network related events and deals with time natively because of the importance of capturing network state over time.

To explore the practical utility of our architecture, we developed a suite of network management applications specifically aimed at enterprise network management. We envision an enterprise network where both basic network functionality as well as sophisticated enterprise specific network management functions are governed by KnowNet-enabled applications. Motivated by a recent report on the topmost enterprise network challenges [8], we developed applications to deal with performance and security management. We show that existing tools can be incorporated into KnowNet, allowing them to interact with new network management functions to realize our objective of holistic network management.

While borrowing from earlier network management ef forts [9], [10], [1], [11], [12], [13], [14], our work on KnowNet is unique along several dimensions. First, we believe our work is among the first to explore the use of a knowledge graph abstraction to capture all of the data, knowledge and actions associated with network management. Second, our approach of attacking the network management problem with a collection of cooperating applications affords a unique tradeoff between flexibility and simplicity. In KnowNet we do not impose structure in our knowledge graph allowing applications to define structure according to the data and management functions they deal with. On the flip side, this requires more sophistication from applications wanting to cooperate. Finally, in KnowNet we take a pragmatic approach to network management (and indeed illustrate the aforementioned flexibility of our approach) by incorporating both existing as well as new network management applications in a holistic framework. This paper makes the following contributions:

• We present the design and implementation of KnowNet, a network management framework that divides the responsibility of reasoning about the network among cooperating applications (§ III).

• We present the design and implementation of Kilo, a knowledge graph that is specialized to the domain of network management (§ III-A).

• To illustrate the utility of KnowNet, we develop a suite of cooperating security- and performance-related network management applications for enterprise networks (§ IV). Some applications in this suite exhibit complex reasoning, while other applications incorporate existing network management tools (sFlow-RT and Snort)—we show how KnowNet enables these existing applications to be brought into a holistic network management system.

• We perform an extensive evaluation of our approach in an emulated enterprise network environment (§ V). This evaluation demonstrates the ability of a KnowNet-enabled network to maintain network health under challenging conditions, and to scale to enterprise-sized networks.