ColonelKernel wrote:Cloud computing represents a new business paradigm too
I completely agree with this statement, but not in the context that follows.
ColonelKernel wrote:Companies are contemplating outsourcing their entire IT infrastructure to "cloud OSes" managed by third parties (Microsoft, Google, Amazon, SalesForce.com) because of the potential cost savings. Whether this is a good idea or not remains to be seen, but it's unlike anything that's happened before in this industry.
How is this unlike mainframes managed by outside firms dialing in for a shell? The only thing the owners or their employees, at many companies, knew how to do was log in from a terminal and run their reports.
ColonelKernel wrote:On the technical side, there is also a big difference: Scale. A mainframe was just one machine with probably tens of users (maybe hundreds). Client/server ala 10-15 years ago would involve maybe hundreds of machines and thousands of users. Cloud computing involves tens of thousands of machines and millions of users.
I agree here, but I did say the paradigm had evolved. Whether it's a couple delivery vans or a fleet of semis pulling triple trailers, we still call them automobiles. Also, it was not always one machine, and one of those machines always had the hardware of 8-64 machines. Mainframes are not multiprocessor, they are multiple full systems from the point of view of the circuitry, with a high-speed local interconnect bus (network) along with, usually, a global supervisory system that can be accessed via serial terminal (or, in slightly more modern systems, telnet/cterm).
ColonelKernel wrote:With tens of thousands of machines, hardware failures become very common. ... The datacenters now have to manage themselves -- automatically taking servers out of rotation, bringing up spares, deploying the right code to them, handle rolling upgrades, etc. All this magic that everyone here seems to take for granted is what goes into a "Cloud OS".
See:
http://en.wikipedia.org/wiki/VMScluster under "Later Developments"
The wikipedia page for IBM's ESA/390(
http://en.wikipedia.org/wiki/IBM_ESA/390) notes that Parallel Sysplex (
http://en.wikipedia.org/wiki/IBM_Parallel_Sysplex) was added in 1994, which allows for the "Coupling Facility" with "Structure Duplexing"(
http://en.wikipedia.org/wiki/Coupling_F ... _Duplexing), which amounts to networked failover between System/390 mainframes. Relatively new, but not Cloud new.
Oh, and can't leave out 1975:
http://en.wikipedia.org/wiki/Tandem_Computers
All of these are smaller scale, but again, I'm talking about repeating patterns in evolution here.