Monday, 14 February 2011

Understanding the Problem

One of the reasons a lot of us work in this industry is because we like to solve problems. We sure aren't in it for the money and fame ;). But how often do you see things suggested or even implemented that indicate a lack of understanding of the real problem?

For example, take the new scalability limits in vSphere 4.1. You can have up to 10,000 VM's managed by a single vCenter - WOW! By why do people actually ask for this? I highly doubt the people who asked for such scalability have licensing concerns - they're likely on a licensing tier that doesn't cost them anymore if they have 1 or 100 vCenter servers. Similarly, removing the need to deploy more hardware is probably a minimal concern - I'm sure a lot of those same people don't even run vCenter on physical hardware anymore. I'm willing to bet the number one reason people asked for higher limits in vCenter was operationally / support driven. But what problem does a single massive vCenter actually address, and what problems does it introduce?

Having a single vCenter means you have a single place to go for provisioning, inventory and monitoring. And that's good, because you don't have to think about it when you need to do one of those things, right? But should you really be hitting vCenter directly for any of those things? I would say not. Your provisioning process should be automated, using some kind of placement engine to determine which vCenter / vDC / Cluster / Datastore / vNetwork to put that next VM on. Your inventory data should be in a CMDB. You should be using some kind of aggregator or at the very least push notifications for monitoring (as opposed to a human "poll").

The problems of a single massive vCenter are obvious - a huge single point of failure, a beefy database, and a nightmare if you ever need to troll through some logs.

So while increasing the scalability of vCenter was the easiest thing VMware could do for customers, it's not really the best thing that customers can do for themselves. Give me multiple small vCenters and decoupled operational tools any day!