Slashdot Mirror
System Administration Cost Studies?
davodavodavo asks: "I'm a market researacher looking for any industry data on the costs of administering UNIX, Linux and Wintel server farms. I've got the overview stuff (Gartner grp), but I'm looking to build a fairly detailed cost model. Specific focus are the activities of server configuration management -- binary images, software distribution, and application deployment. How frequently does some part of a server's software stack have to be updated / upgraded? Any data along the following lines will be appreciated: amount of time it takes an admin to update 50 machine; frequency of distribution or configuration error requiring a roll-back to a previous configuration; average time to perform a rollback; and so forth. If anyone knows of a good detailed model on this topic, please send pointers! I will happily provide the results of my work to the slashdot community (if you are interested, please email me). Basically, I am simply trying to understand the economics behind server management."
Not sure about formal studies but in general I have found that debian when installed with apt-cron and pointing the sources.list to an internal machine with custom packages. That way there is no difference between updateing 1 client and 100 clients.
Don't forget about education costs... Figure that a sysadmin to stay current with technology needs about 4 weeks of training each year... At ~2500 a week that's about 10K right off the top.
Not sure what level of system administration you are looking for. Enterprise system administration is different from a medium or small size company.
It's would probably be better to try to find some complete IT budgets and see what they spend on systems administration. Also look at the metrics for each system type.
/* TODO: Spawn child process, interest child in technology, have child write a new sig */
Forget Gartner, IDC and all those other companies that do NOT DO RESEARCH and just post reports showing whatever it is they want to show (usually that windows is the One True Way).
To figure this out yourself, you're going to have to do direct research that is specific to your environment. Price the hardware. Figure the average lifespans (for a PC its 2 years, for a Mac, 4 years). Look at the OS costs. Figure out directly the support costs-- eg: an average of 4 incidents per cleint seat per week for Linux, 3 for windows and 1 for Mac. The costs of basic training classes, or hiring a trainer (you can call the trainers and ask them how long it will take and how much it will cost) and all that.
In the end you'll find Macs are cheaper- but don't take my word for it (or garnters or anyone elses) do the research yourself.
Its the only way to be sure.
Yeah, and you guys panned the ipod too: http://apple.slashdot.org/article.pl?sid=01/10/23
The Practice of System and Network Admin (well we all know this one, come one on , we ok ISBN 0201702711), has quite of the first two chapters on this. More of why you should have common automatic installs than the actual ROI costs for doing the automatic scripts etc.
BUT ya gotta remember that you'll need more and bigger M$ servers to accomplish the same things with Linux, *BSD or Unix. Hence the cost goes up.
Any data along the following lines will be appreciated: amount of time it takes an admin to update 50 machine; frequency of distribution or configuration error requiring a roll-back to a previous configuration; average time to perform a rollback; and so forth.
I think all and all, the answer is "it depends". The question is not very much unlike, "I am looking for statistical information on how long it takes to repair a car." Of course, repairing a car could mean anything. Putting in a new engine. Changing the spark plugs. It is hopelessly open-ended.
Much in the same way, "updating 50 machines" is also convoluted. We have a seasoned script at work which we use to update a very large number of boxes all at once. It can load software or change parameters, or whatever. The time it takes is usually far less than 1 minute per box.
On the other hand, upgrading the OS can be far trickier. There is the planning stage (working with the application groups to make sure their applications all support the new version of the OS and what patches they need), the implementation stage (which is slow and drawn out, and you have your choices of upgrading the existing machine or installing a new OS on top of it, both with advantages and disadvantages). The implementation usually takes 4 or 5 hours per machine, and is not done in an assembly-line fashion.
Of course, the second instance (os upgrade/replacement) is a case study for Sun's N1. If their software can magically track dependancies in OS upgrades and do it automatically, that'd be swell.