Increasing complexity will render traditional monitoring obsolete
Congratulations! You’ve successfully virtualized your network/server/storage environments and saved xx% (insert percentage). But who besides your CFO really cares? All you’ve done is lower the cost of an increasingly commoditized service, and anyone who’s spent a day working for a vendor knows that’s a very uncomfortable place to be. Unless you’ve also dramatically improved the service delivered to your end users…providing service how they want it, when they want it, and with the ability to transparently adapt to fluctuating demand…the only thing you’ve accomplished is to make it easier for an outsourcing provider to come in and take over your job.
I was reading Art Wittmann’s Practical Analysis the other day and he made several great points – successful vendors focus on user perception of service as much as implementing whiz-bang technologies to make their delivery more efficient. He also describes how tried and true Service Assurance disciplines can work wonders to improve that perception.
But he focused on the event; the provisioning of a new service, the follow-up on an outage. Not that those aren’t good places to focus (painful in most places), but it misses the work that goes on the other 90% of the time – day to day operations that, if done well, are completely invisible to the end user.
In today’s increasingly complex IT environment CIO’s must invest in real-time monitoring, metering and logging that incorporates a performance and health discipline. If you’ve been around long enough to recognize this as a drumbeat from 10 years ago during the dot.com expansion you’re partially right, but there are some key differences. Take an example of a simple transactional application – back then you had some web servers, some application servers, and a database… if you were really sophisticated a link to some legacy system to check inventory before taking an order. It was ‘clean’ – in your datacenter, or hosted entirely by an external provider. Compare that application today: probably a mashup with services from external security providers, checking inventory from a handful of suppliers, visibility into the fulfillment process, and real-time delivery information from the shipper of your choice. Where are those servers, the data stores, and what’s the network that connects them all? Can you detect problems before hearing from users? Better yet, how about a slowdown…do you know where to look?
The point I am making is that traditional ESM approaches are no longer sufficient – applications must be instrumented independently of the physical resources they are running on. Operations teams need to understand true service expectations and the business policies/priority of multiple applications. They must have the tools to help them detect deviations from normal behavior, and the expertise to understand if that deviation is a business change or the symptom of a problem. And ideally they need dynamic execution management tools (informed by the monitors) to shift application resource assignments before a user notices the problem.
Think I’m wrong? I’d love to hear your plan for managing applications in the Cloud…
Jim
