Aaron Huslage

The relative youth of cloud services makes them susceptible to being used for things that they aren’t designed for and, probably, should never be used to do. Of note is this recent entry on Amazon’s EC2 message boards. Richard Haas points out what he sees as a detriment of the service:

What if you want to host highly configured persistent system instances?  Imaging a work group server — every other day you may add cron tasks, additional SMB shares, more Postfix configuration commands, etc.  The “thing” that you care about is the system itself, not some separate volume of data.  The system itself and its capabilities/functions continue to evolve daily, perhaps by the work of multiple administrators across the globe.

You’d like to retain that information/configuration/function across multiple start/reboot cycles … how do you do that?

Right now, the only method that is obvious (to me) is performing instance backups … then constructing new instances from the backups — boy that’s labor intensive and fraught with peril.

Or is EC2 only useful where the system configuration is fixed/generic/non-persistent?  That would IMO be a waste, given the disaster recovery/continuous operation capabilities of cloud computing.

I know nothing about Richard Haas, his motives or his understanding of IT issues, but I see this as a naive view of how the cloud is supposed to function. I think he sort of understands the idea, but the execution has been missed in his mind.

This is, unfortunately, the predominant wisdom of most IT people out there. You set up a server to do some task, like serving files to locally connected users, and then that box runs forever until you get a new one. This model doesn’t work exactly the same way for clouds. Clouds aren’t designed or meant to serve data like that. They are meant to be generic server containers attached to some data storage that contains everything they need to do their job.

Enterprise IT has long held the notion that the data and the server OS are somehow tied together. They are not. The IIS web server is no more a part of of the OS than is MS Word. Apache may come as a package on most Linux systems, but that doesn’t make it any less of an application.

The heart of cloud computing is returning to the model that applications, data and operating systems are completely separate beasts that should not be intimately intertwined. Any one server or application or dataset is completely independent and can live anywhere in the cloud.

Gone are the days of the server request for a big Sun V890 server with gobs of RAM and Terabytes of expensive storage. Those things simply don’t matter in the cloud world. What matters is what application is running, what services it needs, how its data is stored and how it can scale horizontally across many server instances. We have to think more like mainframe people than open systems people.

Enterprises will not be able to take fill advantage of cloud computing services as long as they see them as simple replacements for their current infrastructure. The new model is an update and resurrection of an old one of centralized mainframe computing. Tell me your workload and I’ll make sure the thing keeps running…no matter what physical server it’s on.