Friday, April 24, 2009

Infrastructure Agility via Cloud Technology

I'm honored to have just been published on Edgewater's public blog...

It's bit about managing infrastructure agility. The basic idea is architecting your infrastructure so that you can push off parts to different clouds when you need to, for any of a multitude of reasons. The idea goes a bit beyond virtualization.

Check it out:

http://edgewatertech.wordpress.com/2009/04/24/best-practice-cloud-computing/

Workstation Virtualization

I've been having fun (ya, really... fun!) with MS VPC 2007 SP1 lately.

I've put some VM's on a portable USB disk. USB2 isn't the best connection, but it's workable. It does a few things... adds a spindle to the system config, offloading that overhead from the main disk.

An external disk also means the VM is "portable". I can launch the VM on my laptop or any other host system (like my home system) without any significant difficulty. Even better, by having the VPC config files on the host, rather than on the external disk, you can tune VM settings (like memory and network connectivity) for optimal conditions on the host system.

One other nice time-saver is differencing disks. You can create a "base" virtual hard disk, and create VHD's that are deltas of the base... by doing this, you can create a primary configuration, and then create several machines that inherit that basic config. It came in very handy for a recent product evaluation... I just created a base VM roughly according to what the client expects to host the system on, and then created VMs based on that for each product I wanted to evaluate.

Another nice feature is virtual assist hardware. At first, I didn't know my ThinkPad had it, but it turns out to be a BIOS setting. Flip that on, do a cold boot, and VM performance is visibly better. I knew of some of the other features from past experience with MS VPC 2005, but the hardware acceleration is new to me. (Ironically, my newer home desktop, a 64-bit monster with huge RAM doesn't support the hardware assist... performance isn't a problem, there, tho. )

One more trick: enable the Undo disk option... It put another layer of protection on your VM, allowing you snap a line on your VM at a point in time that you can back-out to. The cool part about this is that the Undo disk is created as a temporary file on the host system, (typically on the primary system drive). This distributes load across the spindles even more, which further improves run-time performance. The downside: when it comes time to commit the undo, it can take a while.

I still love the idea, going forward, of putting client dev environments on a config like this... Not only does it create a nice level of separation between client system configurations, but when you get your hands on better hardware, migration is not an issue.