Why does server provisioning time vary so much?

Hi Max,

I think I have a pretty good answer to your question (though 4 days is never excusable, unless they buy a new server and install it each time a customer orders a box).

LinMin sells software to hosting companies to automate server provisioning and recovery (LinMin Bare Metal Provisioning), so I'm speaking from the experience of having interacted with dozens of hosting companies, but not from that of a hosting company.

- What OS/stack you decide to have installed makes a slight difference, but not enough to justify day+ delays. Debian Linux should take 10 minutes, Red Hat or CentOS 15, Windows Server (2003 or 2008) takes 30 to 40 minutes (including 3 reboots for w2k8). Some cloud companies use our imaging (vs. provisioning) to clone new boxes with identical hardware, and that takes only 3-5 minutes (but you have to have pretty homogeneous hardware for imaging).

I see the differences in sophistication of infrastructure management. From slowest to quickest turnaround time (assuming servers are available, racked and cabled):

- Slow: Send Joe with a DVD to install an OS for Max. Joe gets there when he can (often, the data center isn't near the office). He enters your information by hand (he printed your request to take with him). Reports back when done, and Max gets notified (24 hours isn't uncommon).

- Good: Joe gets notified of Max's request. He goes to the LinMin GUI (or that of another product he's using), enters the MAC address of a free server and IP address, and copies and pastes your selections (OS, host name, time zone, password, etc.), then remotely powers up the system and LinMin Bare Metal does the server provisioning (in 10 to 40 minutes). Here, the only unpredictable delay is the amount of time it takes Joe to see your request and enter the data. Joe doesn't need to be in the data center.

- Best: The fastest way is to have a direct connection (API, or application programming interface) between the ordering system that Max used and the LinMin (or over vendor's) system. If so implemented, the hosting vendor's orchestration software takes Max's request details, plucks a free server with its MAC address from a pool, plucks an IP from another pool and issues within a second or 2 a call to the LinMin (or other vendor's) provisioning server and remotely powers on the system (and LinMin immediately starts the server provisioning process). In this case, the delay is minimal, and provisioning time is truly proportional to the amount of time it takes to install the OS, not travel time, or "I'll get to it" time.

I hope this helps!

LKG