Back in the late 90's I began studying the answer to this question. The research involved reviewing hundreds of products from many organizations. From universities, from open source / free developers, on to the biggest vendors in the market.

What I have come to realize is that an operations management platform must exhibit these key attributes:

• Interoperability
  The typical approach of management vendors has been to create a new API, a new framework, a new management bus, or something along those lines and impose this on the customer. The marketing gimmick being, "switch to our framework and you'll enjoy management bliss!" The problem is that the switch almost always means you lose the ability to manage some device or application you have in your environment. This is not to say that typical management vendors have been trying to do this kind of thing intentionally to their customers (although some might disagree), but rather, they often have an alternative agenda. I.e., they have a platform, operating system, or hardware they need to support and the management model is used to push that agenda rather than to solve the management problem for the customer. The right approach is not to impose a management framework, API, or protocol on the customer, but to build a product that supports the existing management frameworks that a customer has deployed, whether free, public, private, or proprietary.
   
• Flexibility
  The software must be flexible enough to tie into existing systems, to provide different views to every user, to customize the display for customers, to generate customized reports and to adaptively solve the problem of management for the customer. Easy to use "pre-canned" products are great, until you need to do somehting custom with them. The software be as easy to use as ready to go pre-canned software but must also be flexible enough so that the customer can do all of the things they need to do.
    
• Ease of Use
  Management applications are notoriously complex to use, install, and maintain. I remember the story of one early adopter of AutoNOC in Texas. The data center down the street had done the "responsible" thing and spent several million dollars on an installation of a large SNMP based framework (I'll be nice and not mention any names). The vendor had flown in six consultants to build the network model and install the software. The consultants did not really understand the customers network and In the process of discovering the network, they crashed the network because the consultants didn't understand the network. The lesson here is that customers understand their networks far better than external consultants and any workable platform must not require consulting. It must be easy enough for the customer to deploy themselves. The customer will always build a better model of their network than a consultant, but the software must be easy enough to allow them to do this on their own.
   
• Infinite Histories
  Thirty seconds? Twenty-four hours? Sorry, that is not enough data to fully understand how the network is working. Expensive and hard to maintain SQL servers are not going to be able to keep up with the volume of data a proper network operations platform generates. Event-centric systems are not optimized to manage the massive volume of performance data that is required for success. Fully understanding what is happening on a network is only possible when you can go back in time to six months ago, or a year ago and see the real high resolution data. Proper management and network analysis requires the ability to store a near infinite amount of high resolution data.
   
• Scalability
  A scalable operations platform means not only that the software grows with the size of the network, but it is easy to grow the network model. Centralized models lose resolution and exponentially increase bandwidth requirements the closer to the edge that you get. Distributed networks are severable and independently managed. Which is right? The answer is neither is right. Real world networks mix a centralized and a distributed management model. The software must support both types well to be the right product. Events and root cause must come together at some point for global analysis while the model must be fully severable to a degree. The Internet itself follows the partly centralized, partly distributed model. The right management platform will work well in both a distributed and a centralized management paradigm. We think of this as a hybrid scalability model.
   
• Reliability
  The management platform is the last and final wall insuring network availability. It simply must be reliable. If your applications or network goes down, the management platform must not. Maximizing reliability must be a key component of every design decision in the technology.
   
• Performance
  A good management solution has to run on rails. High performance operations management is a more sophisticated problem than even the most powerful databases. Databases have specific data request queries and actions to respond to and that is the only thing they do. An operations management solution has to manage all of the devices, all of the users, perform report generation, query requests, everything, all asynchronously and it has to do it in a fully live, real-time environment. High performance operations management must leverage the latest in computing technologies to maximize the quality of the end user experience.

With all of these requirements on operations vendors, it is easy to understand why the perfect solution has not emerged until today.

Our focus on balancing these criteria from the very beginning has led to an unmatched architectural and design understanding of everything necessary to deliver the optimal management experience for the customer.

It is this focus that has led to the latest release of our core operations platform, AutoNOC 2.2.

Kyle Lussier
President and Architect of AutoNOC