Harvey, did you notice this piece at Anand?
The State of Corporate IT: A case for Linux anandtech.com
It's all sort of old news for long time Microsoft watchers like us, but still sort of breathtaking. A bit from the middle:
Originally, the licensing agreement called for a $20 per seat annual fee for each client that would be accessing NT servers. Originally, two NT multiprocessor NT servers were put in place, each hosting Mail, Internet, File and Print requests. Costs for each server were based upon the number of clients accessing those servers at any one time. Since the company had their employees in separate shifts, only one half of their total employees would be capable of using those servers at any given time. After an initial analysis, it was determined that at no time did the number of concurrent users exceed 2,000. The drafted licensing agreement called for 2,000 concurrent licenses at $20 per annum, for a total of $40,000.
Over the next few years, as the transition from the old server software to the new became complete, changes to the licensing and service agreements were introduced. Uptime over those first years was not nearly as much as the service agreement had called for, and the cost to the company was becoming severe. When tim e came to renew the service agreement, the company sought more assurances and tighter uptime requirements. In response, Microsoft cited increasing demand on the servers and indicated that in order to ensure proper service and support, each server would have to be limited to one of three primary functions: Internet/Mail, File Serving and Print Serving. NT 4.0 performed best when each of those tasks was handled by a dedicated server and system integrity could only be guaranteed if the functions were separate.
In addition, the case was made that with the split of responsibility to multiple servers, the number of concurrent users could not be accurately determined and that it would be necessary to pay licensing fees for all of the clients that might use those servers. Employment at the company had increased by some 1,000 workers and network usage had increased along with it. When all was said and done, the company was asked to expand from two to six servers and to pay client access fees for a full 5,000 users on each of the three primary servers.
Under this proposal, annual licensing fees would increase from $40,000 to $300,000, but uptime performance would be guaranteed at a specific rate and there would be rebates should those rates not be met on a consistent basis. The company had invested over a million dollars to make the switch from the old to the new, and at this point, going back was not a viable option. Reluctantly the agreement was made and they moved forward.
And that was just the beginning, of course. Meanwhile, I'm building up to this retro project of trying to run a win9x system out of a ramdisk. At current prices of a little over $50 a gigabyte for cheap SDRAM, ram is probably about 1/4 the cost per byte that disk was in the early days of Win95. Sucker oughta run like a bat outa hell, and I figure that if you only write the ramdisk back after major upgrades, it ought to be relatively immune to the normal Win9x bitrot, and it'd be easy as could be to back out to a previous version. Plus, the Win9x nonsense of non-reentrant synchronous disk i/o at the OS level wouldn't be a problem anymore if it was all in ram.
What would be really cool is if somebody would build a mobo with enough flash to dump a floppy image into it (oops, most probably have that already), plus the bios hooks to boot of of the floppy image. Then, you could probably set the system up to load the system ramdisk image off a network, and have a pure solid state PC, no moving parts except for the fans. |
