Technology is evolving at a dizzying pace. With that evolution comes a new set of dynamics that affect pretty much every aspect of the traditional information technology (IT) life cycle. Understanding how these dynamics interact can be challenging, and every environment presents its own set of variables that alter those interactions – producing a different “economic footprint” in each case.
This post takes a look at some of the technological, and in some instances business trends that are contributing to the new era of economic factors that make up the cost structure of delivering IT services to customers – whether they are market facing or internal lines of business. It is not the intent of this post to present a formula or “cookbook” for determining the economic footprint of any particular technology or how it contributes to and interacts with the complete IT services delivery ecosystem. Rather, it will hopefully offer some interesting perspectives on how new technology is changing the shape of IT services and possibly generate creative thinking on how to measure its impact.
There are many, many factors that make up any given IT service delivery infrastructure. Far too many to cover in a single writing. This post presents a few that are generating quite a bit of interest, or hype in some cases as we commonly know it. Let’s take a look at them in more detail.
Moore and Metcalfe
What analysis or insight on the progression of technology would be complete without paying the obligatory, and in reality gratuitous homage to Moore’s Law? So in that respect this viewpoint will be no different. The predictions of Moore’s Law continue to hold true and are definitely a factor in the evolution of the makeup of infrastructure models. But volumes have been written on this subject and I think there’s little to be added here other than saying that in addition to Moore’s Law we should take a look at Metcalfe’s Law.
I’m not suggesting that we look at Metcalfe’s Law in the context of its original definition which deals directly with the value of telecommunications networks, but more in the sense of how the number of connections of “services-based elements” in the cloud computing world adds both complexity and value to the new era of information technology. The architectural model of IT systems has morphed from the traditional “north-south” model to an “east-west” model, which has driven significant changes in the design of data centers. Now we are moving to a “north-east-south-west” model, where many of the elements that make up the traditional application ecosystem may be scattered across not only LANS within a data center, but WANS connecting multiple disparate cloud services. This complexity of relationships is driving a new, or at least a renewal of, interest in and energy around the development of systems, both hardware and software, that specialize in establishing and maintaining connectivity across a range of physical and virtual service elements. With complexity comes costs, and in this case these costs are often unanticipated.
Low power, yet powerful
For the last couple of decades the performance improvement of microprocessor technology has followed the path least resistance – simply cranking up the clock frequency to wring out a few more instructions per second. The past few years have shown us the value of stamping out multiple cores on a single piece of silicon which has had a fairly significant impact on the price/performance curve of microprocessors. But we are quickly reaching the limits where tweaking the clock speed provides enough value when compared to the accompanying increase in power consumption and heat generation.
Enter the era of low power chips. Somewhat ironic that we would think in terms of “less is better”, but when you look at the full picture of microprocessor economics, it definitely makes sense to balance all the factors mentioned above. So welcome the likes of ARM, Atom and Tilera. Technology that was once “embedded” inside specialized systems and mobile devices is now making its way into mainstream server technology, offering equivalent if not better performance and very attractive environmental characteristics. Lower power, high density (core-wise) technology is not really new. For several years Azul Systems has offered a very powerful multi-core solution (Vega) for high performance Java applications. But it is just recently that we are seeing traditional servers (and server system components) being built around the lower power processors with new entrants like Calxeda, SeaMicro, Applied Micro and Tilera.
Now this flurry of activity is not validation that lower power, highly dense systems will be appropriate for all workload models. Early implementations have focused on highly distributed, lower core count workloads (e.g., key value stores in distributed hash tables) where “brute force” processing power is not essential. But as new system models and form factors (e.g., Sea Micro’s 768 Atom CPUs in 10RU) emerge, these higher density, lower power systems hold the potential to significantly impact data center “per square foot” economics.
The rack is the new “U”
Going hand-in-hand with emergence of lower power, higher density processors is the notion that servers are no longer acquired on a “per U” basis. We all marveled at the 1U form factor and how we could cram up to 42 of those little pizza boxes in a single rack (Cobalt Systems was an early leader in this practice). But as we know, the physical aspects and management requirements of the connectivity for these razor-thin form factors becomes an issue.
For most mid to large enterprises, and especially for web-scale companies, the hardware acquisition model (at least for servers) is shifting to “rack quantities”. That is, it’s much easier and often more cost effective to buy in quantities of pre-configured racks as opposed to individual units. And with the acceptance of virtualization as an operating norm, it is now much easier to simply buy in bulk so to speak and the parse out individual compute, network and storage elements in a virtualized footprint. This significantly simplifies the planning, purchasing, shipping, installation and physical management costs associated with expanding existing or building out new data center environments. And it offers new capabilities in terms of designing and implementing HVAC (heating, ventilation and cooling) solutions such as “cold aisles” since the thermodynamic characteristics of common components care well known.
The container is the new rack, and it’s all about the PUE
For the really big data centers, racks are a thing of the past. While they still exist, they are now delivered at “container-scale”. Amazon, Facebook, Microsoft, and Google exemplify early adopters of this approach, which was pioneered by Sun Microsystems (Project BlackBox back in 2006, later know as Sun Modular Datacenter). While containerized data centers are pretty much the domain of the web-scale providers mentioned above, we are seeing a few large enterprises such as ExxonMobil, Pacific Gas and Electric Company, E*TRADE, and Toyota adopt this model.
One of the benefits of this container-scale solution is that it allows for the planning, design and implementation of very efficient cooling and power solutions, the latter now measured in terms of Power Usage Effectiveness (PUE), or the ratio of all power supplied to a data center divided by the power consumed by the compute, networking and storage infrastructure. The goal is a PUE of 1.0, where 100% of the data center power goes to running the information technology infrastructure. While not yet achievable, significant progress has been made, with Facebook’s new Pineville, Oregon data center achieving an impressive PUE of 1.07. To reach such a milestone requires a new operational model for data centers, in essence making them “lights out”, which has long been an industry goal. This is now [nearly] possible by the combination of factors mentioned above in conjunction with remote management capabilities and keen insight into the failure rate of technology, which leads us to the next topic.
If it breaks, don’t fix it (until later)
Remember the old days when a server failed? Alarms went off, people scrambled. Sweat trickled down foreheads as IT managers and application owners breathed down the necks of techs and systems administrators as they feverishly worked to bring things back to a normal operational state. Well, those situations still exist, and as we’ve seen lately with Amazon AWS and Microsoft Azure, they have large scale ramifications.
But at the individual infrastructure element level a failure is pretty much a non-event these days. Of course, the failure of a large core router is still of significance, but for individual servers, disk drives, interface cards and similar components it has become pretty much a “let it fail and worry about it later” operational model. With virtualization and failover capabilities, interruptions in a single element of the overall environment have little noticeable impact on system availability or performance. This allows large scale data centers, such as the ones mentioned above, to adopt an operational model that controls the access to the data center floor to only those people who need to find and replace (hot swap) failing components, which then can be repaired and reused as needed. To make this process even more efficient many components in web-scale data centers are being built on a “snap together” design, leading to even lower cost of repair and replacement of failing components.
The data center is the new computer
As we have shown, tight integration is becoming a rapidly increasing norm in data center design. Highly standardized components (servers, storage, function-specific appliances) connected through a mesh-like network architecture is allowing companies to build a unified IT infrastructure model. This massive-scale deployment model in and of itself provides significant economic benefit in its simplicity of design and lower cost of operation (lower PUE, lights-out operation, etc.).
However, the economic benefit of the physical infrastructure has a finite limit. Treating the standardized physical infrastructure as a “pooled” resource significantly extends the benefit in several areas such as acquisition, operational, maintenance and life cycle (refresh) costs. This pooled resource model wouldn’t be possible without two important innovations that have occurred over the past few years – virtualization and the acceptance of self-service as a viable business model. Virtualization is now widely understood and accepted, and while the direct economic benefits are still being developed, it is safe to say that without it the concept of cloud computing would still be an idea scratched on the back of a napkin.
When the scalability, extensibility and reusability of highly integrated, standard infrastructure are combined with the dynamic nature of virtualization and self-service IT, it is clear that defining applications in terms of dedicated physical elements is a thing of the past. We now need to shift the conversation to resource requirements in terms of numbers of users supported, transactions to be processed, or response time – things that are ultimately important to the success or failure of a line of business or market strategy.
The cloud is the new data center
Now that we’ve morphed the traditional data center into a scalable computer that flexes to meet our needs, it’s time to take a look at what the next generation data center will look like. Actually, it won’t be confined by the physical walls of the traditional brick and mortar world. While distributed systems are nothing new, the emergence of the cloud computing model has dramatically extended the boundaries of the traditional application model. As Werner Vogels, Chief Technology Officer of Amazon.com has stated, the future cloud will be a collection of loosely couple services – numbering in the thousands (if not millions) of endpoints which can be connected in numerous ways to create an application ecosystem.
As we in the IT industry are known to do, we have chosen to create different views (names) of what is essentially the same architectural model. Whether it be private, public, hybrid or some other variation, the cloud computing model is characterized by one common feature – the way services are acquired, inter-connected and utilized. That feature in and of itself changes the economics of traditional IT service delivery model. Now I as an end user can acquire IT services without the lag time imposed by traditional data center policies and procedures. This alone holds the potential to accelerate certain aspects of business operations such as time to market and first-mover advantage. But on the other hand, if left unchecked can lead to runaway consumption and a potential “C4” (credit card cloud computing) explosion in IT and line of business budgets – having a detrimental effect on the economics of the IT service delivery model and potentially impacting the performance of the line of business.
Beyond the service aspect, the cloud-as-a-data center model offers some interesting twists on traditional architectures. Companies like Nasuni (cloud storage), Nicira (network virtualization), Okta (identity management) and Nebula (servers) are just a few examples of emerging technologies and capabilities that are leveraging the power of virtualization and the integration cloud models to extend the boundaries of the traditional data center.
Open is in
There is no doubt that the open source movement has been one of the largest contributors to the rapid advancement of technology over the past few years. Leveraging the power of a global community has dramatically improved the IT industry’s ability to bring new capabilities to market at significantly lower price points – creating a powerful ripple effect through all aspects of traditional IT architectures and service delivery models.
Now we are seeing the “open” mindset move beyond the software world into the hardware and operational realms. The Facebook-initiated Open Compute Project (and associated OpenRack specification) is allowing the entire industry to leverage the benefits of its talent and experience in building server technology for web-scale data centers. Open Stack, created jointly by NASA and Rackspace, is building a common cloud operating system that can be utilized across all cloud models – private, public and hybrid. OpenFlow, which grew out of research at Stanford University, is creating architectural models and design templates that enables the separation of network control from the network transport – thus allowing the development of network solutions that meet specific business or technology requirements without modification of the underlying infrastructure. And the Open Data Center Alliance is creating usage models for key data center functions such as secure federation, automation, policy management, service catalogs and standard units of measurement.
The significance of these activities is that they somewhat free IT models from the historical constraints placed upon them by proprietary (vendor) solutions and aren’t burdened by the long and arduous development cycles of traditional standards bodies. The net result is that the industry can now move forward at a more rapid pace and innovate more quickly at all levels – ultimately improving the underlying economics of all IT components and services.
The middleman is out
All of the factors we have discussed up to this point lead us to one last, but significant shift in the IT services industry – the elimination of the “middle man” in the information technology life cycle. With the advent of open architectures for hardware, software, virtualization and data center functions, there is less reliance on “logoed” solutions. Granted, this phenomenon is not universal across all industry segments and is limited to the larger enterprises. But as we have often seen through the trickle-down effect, what happens at the large enterprises and early adopters is eventually enjoyed by the industry as a whole.
An example of this can be seen in mega-scale companies like Facebook who uses a combination of original design manufacturers (ODM) and integrators such as Quanta, Delta Electronics and Synnex to build to spec, integrate, deliver and install at rack or container quantities to Facebook’s data centers. The result of this new paradigm is that traditional big system logos such as IBM, Dell, HP, Oracle and Fujitsu are now being excluded from the supply chain and subsequent upsell opportunities of these very large internet companies and enterprises. As time progresses, this will exert more pressure on those companies to be more competitive across all product lines – changing yet another aspect of the infrastructure economics.
As I stated at the beginning of this post, there are lots of other disruptive trends and technologies (data explosion, mobility, social supply chain, big data, etc.) not covered in this post that also shape the economic footprint of an enterprise’s IT infrastructure and service delivery capabilities. We will save those for other conversations.
One of the ongoing challenges of our industry has been the quantification and qualification of the costs and benefits associated with and derived from IT. Hopefully the sampling of factors covered in this post will provide some insight into how business, technology and service innovation is not only reshaping the economics of the infrastructure that enables IT services, but is also changing the dynamics of the industry as a whole.