Datacenter Network Capacity Management

Here is a complex topic: Capacity Management inside the Datacenter.

There is daily operation and monitoring, and there is capacity management. They will touch themselves in some points but they have completly diferent perspectives:
I have been doing capacity KPIs for the past 8 years. To be honest, I believe that now it is the first time I am doing it right. Here are some diferences:

Operation:
- Realtime monitoring
- Alarm driven
- Focused in detail
- We are worried with uptime

Capacity Management:
- Its a recurrent process
- It is weekly based
- We are worried with provision capacity
- We are worried with long term stability
- We want to have a big picture of all infrastruture.

So, the first questions that normally arise are:
- How can I extract my data
- What are my limits
- Which tools can I use.

Well , the correct questions should be "How can I setup a process that will take actions on important findings triggered by capacity management". If you can answer this question and execute in a process then , you already have capacity management in place. Now, all you need is data and analytics to feed this process. Believe me that setting the process is the hard part because it is normally taken as a less priority task that can always be done later.

So, after the process is setup and running, which means, you will have a weekly review, meeting minutes weekly report that will show you are doing it will all  updated findings, than you will have something for an auditor in a ISO 20000 certification.

But, what you are probably trying to get is, what should I measure and how. Getting the data and finding the limits is also not easy.
If you work in a service provider, you will have dozens of networking device vendores, diferent models, diferent version in the same models (templates), and all will have diferent limits and measurements. Also, the configuration applied to simliar appliances will have diferent measurements because the configuration is diferent. For example, a Cisco ASA configured with multi context must have diferent measurements then a non virtualized chassis.








(will continue ...)