I recently attended a dot-com launch. What was unique about this particular launch was that the company does not own any servers, and they do not have a data center or even rack space in a co-location facility! Instead they run on virtual servers controlled with a Web service interface, and store their data “in the cloud” using a Web service.
Virtualization is almost a cliché in 2007. IDC lowered its server revenue forecast as a direct result of virtualization. Especially because IDC’s report missed the new phenomenon known as “Web Scale Computing”. Traditional virtualization is about sub-slicing your servers into smaller virtual machines. Web Scale computing is about (in part) renting virtual servers from an essentially infinite pool of very inexpensive servers, with the result that fixed costs are eliminated, incremental costs are significantly lower than traditional costs, and capacity planning is an irrelevant exercise.
This isn’t just theory: Federal Computer Week reported in an article published on October 30, 2006 that the Defense Information Systems Agency (DISA) spent $5 to develop a system that would have required a $30,000 server investment if built in-house.
Web Scale Virtualization: Greatly Reduced Cost and Added Flexibilty
Is Web Scale computing this week’s catch phrase, or is it really something new? Industry pundits say that this is a distinct inflection point in the history of computing. Part of the change is technical (XML Web Services and virtualization), and part of the change is a new business model (no fixed costs, and Moore’s law applied to incremental costs).
There are five key things that set Web Scale computing apart:
1. As mentioned above, until now most virtualization focused on slicing your own physical servers into multiple pieces. Web Scale organizations can run entirely “in the cloud”, with no physical vestige of a local data center.
2. A decade ago organizations cut costs by outsourcing their IT operations. Unlike outsourcing, your data center will be virtualized; however your staff will still operate the (now virtual) data center.
3. There are no lower or upper bounds on a Web Scale data center. Adding or subtracting capacity is as simple as making a programmatic request to a Web Service control interface. The result is that you never run out of server capacity, and you never overspend in order to ensure that there is adequate reserve capacity available. Automation allows server farms to expand or contract on an hour-by-hour basis; with the result that you always have “just enough” processing capability. Note that this scaling is only feasible when the underlying pool of servers is so enormous that your requirements are not statistically significant in that larger pool of resources.
4. There are no fixed costs associated with a Web Scale data center. Because all costs are variable, budgets shift from capital expenditures (CapEx) with annual amortization to purely transactional (OpEx) costs. Tax benefits are likely to follow for many commercial enterprises; and even non-commercial organizations will realize dramatic savings.
5. A few world-class operators will provide so much value in the form of operational excellence that it makes no sense for others to even try to run their own operation. This is much the same as the way that FedEx established itself as the de-facto way to deliver packages overnight: no one today would seriously consider operating their own overnight package delivery network on FedEx’s scale.
Abstraction: a Theory of Natural Evolution
Services Oriented Architecture is a form of abstraction. Particularly when implemented as Web Services, architects may have a data-centric view of the business operation they support, with one Web Service that represents a view of inventory, and another data-centric view that exposes – for example – an order-entry system. Applications that consume these services have no knowledge of the underlying details such as database schemas, the operating system, etc. In fact, with a set of loosely-coupled interfaces there is no compelling reason to control most aspects of an application’s physical implementation.
And abstraction is not new to the industry. Runtime environments such as Java and Microsoft’s .NET Framework provide significant benefits in the form of enhanced productivity and security; asking in return that developers relinquish some of their control on under-the-hood knobs and dials such as memory management and thread management.
There’s no reason that abstraction needs to stop with software, though. Sun’s Jonathan Schwartz expressed much the same thinking in a blog posting in October, 2006, titled “The Death of Yesterday’s Datacenter”. He asserts that with today’s network infrastructure we may soon revisit the assumption that each organization needs their own data center.
That’s a valid point. The intersection of Moore’s Law with “World Class Services” is already causing a tidal shift (see the reference above to FedEx). Why invest in a data center and staff when you can rent a server for $0.10 per hour (that’s approximately $72/month), with the ability to add and subtract servers from your virtual “server farm” on an hour-by-hour basis? Especially when these virtual servers are in a physical data center staffed by a world-class operations team.
A Virtual Data Center Needs More Than Just Servers
What does a virtual data center look like from an architectural point of view? As indicated in the illustration, there are three core functions that any physical data center delivers: computing power, data storage, and messaging queuing.
Each of these essential components can be virtualized through the magic of XML Web Services. (Servers are controlled by a Web Service interface instead of being a Web service per-se.) You are free to develop any sort of interface that suits your needs, without restriction on how you innovate. And best of all, XML Web services are well-suited to a SOA architecture.
Is This Right for My Organization?
For many IT executives the salient question is whether the benefits of virtualization outweigh the resulting loss of control. That decision may be purely emotional; and accordingly not every organization will switch. However, given the annual cut-costs-from-IT dance, virtualization’s siren song is seductive.
Whatever you believe, current adoption trends are undeniable. And moving to a virtualized environment starts with a solid SOA foundation. Aren’t you glad you’re ready?
