10 Comments

Summary:

There’s a common assumption that the Cloud’s destiny is to be a public utility. Mark Thiele, of data center operator Switch, argues that would kill competition and innovation, and that IT can be a better option.

CloudUtilityBig
photo: Rani Molla/GigaOM

One of the prevailing assumptions around the cloud computing market is that it will drive towards an über-simplified delivery model that is similar to a utility. Further, this utility model will largely remove the potential for differentiation by most vendors and will lead to a race to the bottom from a pricing perspective.

There is ample evidence commoditization is occurring, and we could point to almost any area of IT to see it, from servers, PCs, virtualization, storage, networking, and so on. However, what is often lost in the obvious is that it’s not that simple.

IT commoditization vs cars

It’s true that with a modern server and chip combination you could likely solve almost any specific workload demand of a modern application. The inherent risk though is that there’s always someone out there looking to make a better rat trap, and the market continues to show that there is real demand for differentiation – consider the viability of both ARM and Intel chips for use with different job types.

For instance, in the case of Intel you could easily brute force the same workload that an ARM chip could handle, but processor by processor you would likely be very inefficient from a utilization and power consumption perspective. The same is true for ARM chips being used where a larger Intel processor might be more effective. While the aforementioned example is a simple one, it applies across most layers of infrastructure: storage, network, I/O, virtualization, and so on.

To make an analogy: Cars have been around for over 100 years now, they must be commodity by now, right? They all (mostly) have four wheels, two or more doors, a combustion engine (mostly) and generally get you from point A to point B successfully.  Are cars a commodity compared to each hand-built car of the late 1800s and early 1900s? Maybe. But only in the sense that we can each buy a Ford Focus with the exact same feature set as another Ford Focus.

You can also buy a Ferrari, which is excellent for country highways and tight corners, or you can pick a Dodge minivan, which is better at hauling the soccer kids around. Each of these two cars have substantially different features and solve different problems, yet they are both cars. So, the simple answer is no, they are not commodity if by commodity you mean there is little or no profit or differentiation to be found.

Drivers that make unique IT solutions critical

CPU Performance – There are some tasks that will need the fastest possible processors. The benefit comes from the time reduction associated with running a workload. The time is so valuable that the cost of the infrastructure and power is immaterial. In many cases this type of environment is refreshed 18-month cycles, but sometimes as few as every six months.

Network Bandwidth – Critical if the data being manipulated or distributed is being moved outside the confines of where the compute resides, or needs to be moved fairly quickly in very large amounts.

Network Performance – In terms of not just bandwidth but also latency. In some cases customer demands are on the level of differences of nanoseconds.

Storage Scale vs. I/O – Similar to networking, the type of storage you need is dependent on the type of work being done. You don’t solve an I/O requirement by just buying bigger arrays with more and larger SATA disks. You also don’t put large cache or expensive memory on storage that is mainly used for archival or lower performance requirements, such as for photos.

These only scratch the surface of the variables associated with building an infrastructure environment, but clearly it would be difficult to create a small handful of solutions or solution providers to take care of every IT workload demand. To the contrary, in my work I regularly see a significant number of players enter the market that either fill an unmet need from an industry perspective or enable new types of performance and pricing models.

Why commodity IT is a bad idea

Not only do I think we’re many years away from having a small handful of service providers deliver us compute on demand, I propose that more importantly, we should all hope that the day never comes anyway. I am against the idea of a utility form of compute delivered to everyone much the same way for many reasons, but the two that I think are the most critical are monopolies and innovation.

Cloud monopolies would be bad

The arguments against service providers acting under the pretext of a public utility are legion: inefficiency, waste, corruption, etc.  (For more detail, refer to the eye-opening book “The Fine Print” by David Cay Johnston.) Considering the importance of compute to the global economy, the last thing we should ever want then is for it to become a monopoly. If we allow a few companies to push the technology to a true commodity business model, then we could count on real competition for service delivery to disappear, as there wouldn’t be an easy way for the little guy or regional player to participate in the market. And it’s the ability for little guys to introduce innovation that sparks competition and evolution.

If Cloud were to become a monopoly service, we would quickly find ourselves suffering the same issues with many other public utilities:  running on older equipment, getting charged for non-existant services or things we don’t understand, and having no one accountable to address complaints.

Innovation would be stifled

Beyond that, Innovation in any market only occurs through necessity. Without real competition, the commodity cloud services would begin to act like so many of the early infrastructure outsource providers did and only deliver to the lowest common denominator. They would make changes more slowly and as customers we would be forced to plan our businesses around what providers were willing to do. IT will be its most successful when the business customer doesn’t have to consider the “limitations” before advancing strategic opportunities.

While there’s little doubt that several big players would come to dominate the market as is already the case today, we cannot afford to be without all those spunky new companies looking to carve out a market for themselves. These little players will force the larger players to stay honest, to bill correctly, to offer new services, and to continually innovate.

The Good News

I don’t really think we have anything to worry about, because as I’ve already indicated I don’t believe we are at any near-term risk of getting to a utility-type delivery of compute resources market. There are just too many ways to (in this case) build that car. So rest easy: We’re going to continue to see lots of great innovation in the infrastructure and application services and delivery space for some time to come, but keep your eyes and ears open all the same.

Mark Thiele is executive VP of Data Center Tech at Switch, the operator of the SuperNAP data center in Las Vegas. Thiele blogs at SwitchScribe and at Data Center Pulse, where he is also president and founder. He can be found on Twitter at @mthiele10.

  1. A number of assumptions here, so let’s go through each …

    “Cars have been around for over 100 years now, they must be commodity by now, right?”

    Here you’re assuming that the process is time based (like diffusion i.e. adoption over time) when in reality it depends upon the actors in the market and competition. Hence evolution from genesis of something to commodity is based upon ubiquity (i.e. how widespread something is, user competition) and certainty (how feature complete something is, i.e. vendor competition).

    The nut and bolt took almost 2,000 years to go from genesis (first creation) to late product and commodity (starting with Maudslay’s screw cutting lathe). Electricity took 1400 years from the Parthian Battery to Westinghouse / Tesla and A/C utility provision. Telephony about 100 years and computing infrastructure about 70 years. So, you cannot assume some linear relationship with time.

    Secondly, in the above examples (Nut and Bolts, Electricity, Computing Infrastructure, Telephony) these systems have become components of something else (e.g. Machines, Consumer Electronics, Big Data System, Smart Phones).

    In industries where the car is a component of something other value chain (e.g. Taxis, Fleets, On Demand hire services etc) then it would be worth looking at whether standardisation has or is occurring (e.g. easy cars etc).

    As you point out the car itself has many commodity components but if the system you’re examining is the top level system then you’re always going to have branding, perception of other values which impact it e.g. Gary Dahl’s “Pet Rock” is probably the best loved example of associating a branding value with what remained a commodity … it’s not a rock (which it was) but a “Pet Rock”.

    Your comparison of computing infrastructure to cars is one of comparing a component to a higher order system which has other values (i.e. personal branding, status symbol etc). If you compare like for like you’d probably end up with a different conclusion i.e. how many users of Amazon EC2 care or even know what hardware Amazon EC2 runs on – it’s an invisible component, far down the value chain. The same can be said of how many users know or care what specification of nuts and bolts are used to build their car?

    “If we allow a few companies to push the technology to a true commodity business model”

    First, it’s not companies that drive things towards a commodity model but the interaction of users (demand competition) and companies (supply competition). The question of evolution (which for activities we call commoditisation) is separate from the question of centralisation / decentralisation and the two shouldn’t be conflated.

    It would have been relatively trivial for the hardware manufacturers to create a price war in the IaaS space around 2008-2010 in order to fragment the market by increasing demand (computing infrastructure is elastic) beyond the capability of one vendor to supply. The fact they didn’t is their own fault and also the reason why we might see centralisation.

    In general,

    * The process of evolution (driven by demand and supply competition) is not time based but depends upon the overall interactions of ALL actors (users and suppliers). It is an inevitable process of competition.

    * The question of centralisation / decentralisation varies with a number of different economic forces but it usually depends upon the actions of SPECIFIC suppliers. Often you will find that companies are suffering from inertia to change (due to past success) and hence new entrants into a market that is commoditising can quickly capture the market. This doesn’t need to be the case though and the issue is one of executive failure usually of past giants.

    “Innovation would be stifled”

    Quite the opposite. There are two aspects of innovation here – operational efficiency and creation of higher order systems.

    The commoditisation of any pre-existing activities which acts as a component leads to rapid increases in the genesis of higher order systems. From Nuts and Bolts to Machines or Electricity to Radio / Hollywood / Consumer Electronics / Computing or Computing to Big Data and all the changes we’re seeing today. Commoditisation always increases innovation (as in genesis of higher order systems), this effect is known as componentisation (Herbert Simon).

    Equally, innovation behind the interface doesn’t stop i.e. electricity was standardised to an interface but innovation still occurs in the means of production (e.g. operational improvements and efficiency or new means of generation). This is the same with all manner of industries from finance to manufacturing (see float glass method of producing glass etc).

    You cannot therefore state that commoditisation inhibits or limits innovation, where historical evidence shows it not only allows for innovation in production but accelerates innovation of higher order systems (i.e. each of the major ages – industrial, mechanical, electrical, internet are associated with commoditisation of pre-existing activities).

    “Drivers that make unique IT solutions critical”

    You’re absolutely spot on that even in a mass market of commodity components there are always niches – same as electricity, finance, most manufacturing etc and you’d expect the same with computing. There will be highly profitable but small niches.

    “There are just too many ways to (in this case) build that car”

    One of the most important parts of any of the common cycles of changes (i.e. ages) is the disruption of past giants stuck behind inertia barriers due to past success. Disruption comes in two forms – there’s the unpredictable, unforeseeable (e.g. a change in characteristic of a product such as disk drives) and then there’s the predictable (e.g. a shift of an activity from product & rental services to commodity and utility services).

    Both cause disruption, the former because it’s unseen, the latter because it’s seen but the company is unable to react (due to inertia which is often embedded in the culture of the organisation). We’ve seen this in every major age and there is nothing which suggest that cloud (which was predicted by Douglas Parkhill in his 1966 book the challenge of the computing utility) will be any different.

    The list of companies who believed that their industry would not be commoditised is long and glorious from the gas lamp companies of the past to almost certainly modern infrastructure companies. The problem for all is that their offering is just a component.

    Share
  2. Simon, as always a well thought out and interesting comment. Time will tell, which one of us is on the right track here. Its likely that it will be 3-5 years before we can be certain the path that “utility” oriented compute is taking. I look forward to continuing the discussion.

    Share
  3. Mark excellent read , totally agree with your points above.
    I think that now with Google, Microsoft, OpenStack, HP, Dell entering the game it becomes even less reasonable to stick to one provider. The other point on that regard is that the cost and effort of building a portable solution that wouldn’t be bound to a particular cloud API is also significantly lower with frameworks such as JClouds and Cloudify as well as Rightscale that provide multi-cloud integration that makes the switch cost pretty low.

    I’ve written a post on the five misconception on cloud portability
    http://natishalom.typepad.com/nati_shaloms_blog/2011/11/five-misconceptions-on-cloud-portability.html

    Share
  4. adrian cockcroft Wednesday, November 21, 2012

    The Fiat-Chrysler company owns both Ferrari and Dodge. AWS has a far broader range of instance type options, services and features than you will find in most datacenters or other clouds. So what’s your point again?

    Share
  5. Adrian, nice flashy answer, but doesn’t change anything. If Amazon and a few others were to “own” cloud services, what would the likely outcome be for service innovation and pricing? The only thing that keeps Amazon innovating in the long run is the effective threat of competition.

    Share
  6. as

    Share
  7. Grids can be commodity at low end. Looking at the Center for High Throughput computing, scientists need to add resources from 0.03 million hours on desktop to 712 million hours acceding the Open Science Grid (40,000 plus servers growing daily) and beyond.

    Problems like “Simulate crop yields and climate change impact at high-resolution (global extents, multi-scale models, multiple crops – corn, soy, wheat, rice)” carried out at University of Chicago Computing Cooperative are not possible to solve using commercial cloud providers, who do not have the scale and the technology for such leading edge application. One interesting observation is that operating on such a large scale, it creates the illusion of infinite resources inside a grid . Note the “illusion of infinite resources” is subjective. In some cases a few computers or AWS instances are sufficient. In HTC, this illusion can only be achieved after accessing many thousands processors at once.

    Share
  8. I read much that compares Cloud to a utility model. I doubt it will become a utility in the sense of a large power or telecom firm. To an extent focusing on the elements of computing today like cpu, memory, bandwidth (infrastructure) I think the real excitement of the Cloud will emerge as the ecosystem of software services in the Cloud grows, and as services currently delivered via hardware migrate to software services. Overall Cloud portends much less focus on infrastructure (although it will certainly change dramatically and continue to matter) and greater focus and larger spend on software services. In the enterprise today in my opinion, beyond the cost of building new systems, the risk associated with building out infrastructure as well as paying up front for software is high, and outcomes vary widely. In a world where infrastructure becomes more stable and easier to provision both the risk and complexity of building software systems decreases. As risk decreases and outcomes/returns become easier to forecast I feel the demand for software will increase dramatically.

    Share
  9. Mark, interesting post, we often come across new cloud customers because they are seeking differentiated service offerings that they are not finding elsewhere. All users must contend with the architecture tradeoffs made by the different IaaS providers, and the limitations inherent in those decisions, and how they relate to the users application workloads. There are a large number of design decisions (tens of thousands) that go into the design of a cloud computing service platform from both a technology and an operational perspective, and those differences drive users to seek alternative solutions. We think this trend will continue because based upon each cloud providers engineering decisions along the way there will be trade offs that users will be sensitive to. We have all seen for instance in the development of switches and routers, that if a manufacturer tries to make a device into a massive “God box” it often is unreliable, suffers from poor performance in many areas, or is overly costly. I think everyone can agree there is commoditization going on at all levels in computing infrastructure. However, it is early days and there is still important work to be done, and significant value to be created. Simon, your comment about how turning certain technologies into commodities can accelerate the creation of “higher order systems” is logically correct, and the cloud ecosystems that will be created by the larger cloud providers over the next few years will likely be impressive in their capabilities. You are also likely correct in that there will continue to be small niches for smaller providers to thrive in. The question is: will the niche players only be succeeding due to their sheer market focus and will they generally be lacking in innovation, and larger players will be the only ones driving the “real” innovation in the industry? We think the smaller cloud providers with deep expertise will continue to innovate for years to come, and that many of the technologies they will develop will end up being acquired by the larger providers and integrated into their cloud ecosystems. It would be unprecedented if companies like Amazon AWS, HP, IBM, etc. never acquired another cloud service provider for its technology. The acquisitions in the beginning will likely be mostly for the software technology, and then we will likely start seeing the larger providers looking to acquire them for their customers too. We continually see dissatisfied cloud users looking for better solutions that are highly optimized to support their specific application workloads, and we do not see this demand going away anytime soon. For those interested, check out our webinar that covers why we think cloud orchestration, bare metal and virtual compute, and advanced networking capabilities are foundational elements that make up a high performance cloud service: http://docs.nephoscale.com/#!/video/53216638. Go straight to the 24 minute mark to see our CTO discuss the platform elements.

    Share
  10. Bruce Templeton Thursday, November 22, 2012

    The previous comment was made by Bruce Templeton, CEO of NephoScale.

    Share

Comments have been disabled for this post