Stateless Servers

  For a few months now, I have been learning all about the stateless server architecture as I will be one of six analysts supporting it when my employer chooses a vendor and purchases the hardware as part of our virtual server and desktop initiative.  Its a great technology that I thought I'd write a little about.

State of the Union a Server
  Let's learn about what stateless servers are by learning what a stateful server is.
What I am referring to when I say "state" is the settings of all the different components of a server.  Some of these are:

• BIOS Settings
• Boot Order
• CPU Settings
• Power Settings
• Etc...
• RAID Settings
• Fiber Channel Adapter Settings
• WWN
• Boot From SAN
• NIC Settings
• Speed/Duplex
• IP Address
• Subnet Mask

  This is a small subset of all the states a server can have.  These settings are stored in various places on the local server and you would have to configure these settings on every server.  If (when?) the server crashes and burns, you must go through the process of restoring the state of the server by reconfiguring all the settings affected by replacing hardware (or entire server!)  Obviously, in a situation like this you need local access to the server (or have an environment where the out-of-band management controller gets network settings automatically assigned (and the "state" of the out-of-band managenent controller itself does not need manually configured!))

Stateless = Simple Management and Recovery
  Now that we know what a stateful server is, we can infer that a stateless server is the opposite.  By moving the ownership of all the configuration points of a server to a centralized location, we can create server (or service) profiles (collection of server settings) for the servers and assign them to the hardware to create a server.  If something happens to the hardware, we can simply move the profile to different hardware and, Presto! the server is back.
  If you haven't guessed by now,  this model works best with blade servers with no local storage (all blades boot from SAN). Keep an extra blade in the chassis you can use if you lose a blade and you can have a failed server back in service in about 10 minutes.


I am sure I am not doing this explanation justice so I will point you to a blog that should make the concept of stateless computing more clear:  http://www.mseanmcgee.com/2010/04/the-state-of-statelessness-cisco-ucs-vs-hp-virtual-connect/

Keep in mind that Cisco really pioneered this with their UCS blade servers.  HP later released Virtual Connect which attempts to compete with UCS but IMO falls short.