Tuesday, October 21, 2014

Cloud systems are the new mainframe

The history of computers can be divided (somewhat arbitrarily) into six periods. These are:

Mainframe
Timeshare (on mainframes)
Minicomputers
Desktop computers (includes pre-PC microcomputers, workstations, and laptops)
Servers and networked desktops
Mobile devices (phones and tablets)

I was going to add 'cloud systems' to the list as a seventh period, but I got to thinking.

My six arbitrary periods of computing show definite trends. The first trend is size: computers became physically smaller in each successive period. Mainframe computers were (and are) large systems that occupy rooms. Minicomputers were the sizes of refrigerators. Desktop computers fit on (or under) a desk. Mobile devices are small enough to carry in a shirt pocket.

The next trend is cost. Each successive period has a lower cost than the previous one. Mainframes cost in the hundreds of thousands of dollars. Minicomputers in the tens of thousands. Desktop computers were typically under $3000 (although some did edge up near $10,000) and today are usually under $1000. Mobile device costs range from $50 to $500.

The third trend is administrative effort or "load". Mainframes needed a team of well-trained attendants. Minicomputers needed one knowledgeable person to act as "system operator" or "sysop". Desktop computers could be administered by a geeky person in the home, or for large offices a team of support persons (but less than one support person per PC). Mobile devices need... no one. (Well, technically they are administered by the tribal chieftains: Apple, Google, or Microsoft.)

Cloud systems defy these trends.

By "cloud systems", I mean the cloud services that are offered by Amazon.com, Microsoft, Google, and others. I am including all of the services: infrastructure as a service, platform as a service, software as a service, machine images, queue systems, compute engines, storage engines, web servers... the whole kaboodle.

Cloud systems are large and expensive. They also tend to be limited in number, perhaps because they are large and expensive. They also have a sizable team of attendants. Cloud systems are complex and a large team is needed to keep everything running.

Cloud systems are much like mainframe computers.

The cloud services that are offered by vendors are much like the timesharing services offered by mainframe owners. With timesharing, customers could buy just as much computing time as they needed. Sound familiar? It's the model used by cloud computing.

We have, with cloud computing, returned to the mainframe era. This period has many similarities with the mainframe period. Mainframes were large, expensive to own, complex, and expensive to operate. Cloud systems are the same. The early mainframe period saw a number of competitors: IBM, NCR, CDC, Burroughs, Honeywell, and Univac, to name a few. Today we see competition between Amazon.com, Microsoft, Google, and others (including IBM).

Perhaps my "periods of computing history" is not so much a linear list as a cycle. Perhaps we are about to go "around" again, starting with the mainframe (or cloud) stage of expensive systems and evolve forward. What can we expect?

The mainframe period can be divided into two subperiods: before the System/360 and after. Before the IBM System/360, there was competition between companies and different designs. After the IBM System/360, companies standardized on that architecture. The System/360 design is still visible in mainframes of today.

An equivalent action in cloud systems would be the standardization of a cloud architecture. Perhaps the Open Stack software, perhaps Microsoft's Azure. I do not know which it will be. The key is for companies to standardize on one architecture. If it is a proprietary architecture, then that architecture's vendor is elevated to the role of industry leader, as IBM was with the System/360 (and later System/370) mainframes.

While companies are busy modifying their systems to conform to the industry standard platform, innovators develop technologies that allow for smaller versions. In the 1960s and 1970s, vendors introduced minicomputers. These were smaller than mainframes, less expensive, and easier to operate. For cloud systems, the equivalent would be... smaller than mainframe clouds, less expensive, and easier to operate. They would be less sophisticated than mainframe clouds, but "mini clouds" would still be useful.

In the late 1970s, technology advances lead to the microcomputer which could be purchased and used by a single person. As with mainframe computers, there were a variety of competing standards. After IBM introduced the Personal Computer, businesses (and individuals) elevated it to industry standard. Equivalent events in cloud would mean the development of individual-sized cloud systems, small enough to be purchased by a single person.

The 1980s saw the rise of desktop computers. The 1990s saw the rise of networked computers, desktop and server. An equivalent for cloud would be connecting cloud systems to one another. Somehow I think this "inter-cloud connection" will occur earlier, perhaps in the "mini cloud" period. We already have the network hardware and protocols in place. Connecting cloud systems will probably require some high-level protocols, and maybe faster connections, but the work should be minimal.

I'm still thinking of adding "cloud systems" to my list of computing periods. But I'm pretty sure that it won't be the last entry.

No comments: