Following the interest at my Certification Quickstart sessions at vForum Sydney I am doing a series of blog posts about studying for the VCP5 exam. One of the big things I believe you need to become a VCP (and good at working with vSphere, which is even more important) is a test lab. The test lab is a place where you can make and break every feature of vSphere without having to ask permission. In a previous post I looked at using a PC with VMware hosted virtualisation as a lab.
This blog post is about the most obvious lab, a set of retired servers. ESX has always been happiest running on server class hardware and it is the best representation of the hardware that is used in production. A lot of customers retire their physical servers after three or four years service, Quad core CPUs are being retired now, early Nehalem CPUs can’t be far away from retirement. These servers typically have multiple NICs and a fair amount of RAM, they may even have Fibre Channel and if you’re lucky there may even be a retired SAN.
The big downside to retired servers is that they are servers. They are noisy, they consume a lot of power and they generate a lot of heat. Running these in an apartment won’t impress your family and the cost of power will be high. Maybe retired servers are a better lab for leaving in the office. My retired servers are a pair of HP blades in a blade enclosure, they live in a shed. Sometimes I can hear them from inside the house, although they don’t keep me awake the way they did before they were moved to the shed. Air flow through the shed (or air-conditioning!) is important. Unlike a PC, a good server will power itself off if it gets too hot.
A good thing about servers is support for a lot of RAM, 8 DIMM slots and support for 8GB DIMMs. I currently have 28GB between the two servers, this allows me to run a Windows SBS 2008 environment as well as vCentre and a full View 5.0 and a few drone VMs and virtual appliances. When the RAM doubling upgrade arrives I will also be able to run some nested ESX on top for SRM 5.0 and maybe VC Operations when the new version ships.
Two retired servers isn’t really enough, you also need shared storage. My shared storage is a currently a NetGear ReadyNAS. This provides NFS datastores which are the performance bottleneck in this lab, 4 spindles of SATA using RAID5 means lots of space but poor write performance. If your retired servers came with a lot of spindles of local disk you’re in luck, for a lab you can run Openfiler in a VM and make that local storage shared. For a lab it doesn’t matter that the storage is dependent on one ESX server, just make sure you power off the other host when you’re testing HA. If you have three or more servers with local disk then you could deploy the VSA to two of the servers (the other one will have to run vCentre, preferably in a VM). The VSA is something you should be learning about too.
If you’re extremely lucky and you also got a retired SAN then you definitely want to leave this in the office, SAN consume a lot of power and therefore produce a lot of heat and noise. The benefit is that you can only really learn Fibre Channel by using Fibre Channel, along with the fact that more spindles of disk means better performance for the VMs.
A test lab made of retired servers, particularly with a Fibre Channel SAN, is something that will get left running all the time, startup and shutdown will each take most of an hour. As a shared lab in an office where there’s plenty of power and cooling available and where the Wife Acceptance Factor (WAF) is not relevant then they are great. In an apartment or inside the house I’d suggest other options, my next post will be about using a whitebox ESX server.
© 2011, Alastair. All rights reserved.