A look at the current-state cloud with a perspective on performance
Think PCs in 1982. The Ford Model T in 1908. Or go all the way back to the Watt steam engine in the late 1700s, considered a major catalyst of the industrial revolution. Tech watchers are likening the rise of cloud computing to any or all of these paradigm-changing milestones.
There's much evidence to back up such analogies, if current trends hold, as they certainly appear to be. Consider:
Every consumer with a smartphone, every Web user with a Gmail or Hotmail address, every Facebook friend is actively engaged in the cloud.
The U.S. government is moving hundreds of its data centers to the cloud. The Pentagon has multiple cloud initiatives underway that will ultimately touch every soldier in the field, and has already equipped Special Operations Forces with Android-powered tablets. 1Wired.com, "Pentagon Looks to Militarize the Cloud," by Spencer Ackerman, 2/22/11 NATO is piloting an IBM-built system at its Norfolk, VA facility, and hopes to roll it out to all 28 alliance nations. 2The Wall Street Journal, “NATO Starts Cloud Computing,” 12/22/10
In the business world, adoption of software as a service jumped from 66 percent of companies in 2007 to 90 percent in 2010. 3Microsoft White Paper, “The Economics of the Cloud,” 11/10, quoting Gartner data
On the provider side, big companies are snapping up smaller data storage companies with fervor reminiscent of the dot-com boom. Verizon this winter announced a $1.4 billion deal to acquire Terremark Worldwide. And Hewlett-Packard, EMC, Dell, and IBM have all ponied up to buy "a slew" of data storage companies, according to a New York Times report. 4The New York Times, DealBook, “Verizon to Buy Terremark for $1.4 Billion,” 1/27/11
The tech giants are all investing huge sums in their own data centers. Microsoft CEO Steve Ballmer announced last fall that the company will invest billions in cloud data centers. 5Bloomberg BusinessWeek, “Microsoft’s Ballmer to Invest Billions in Cloud Data Centers,” 10/6/10. Apple is opening its billion-dollar data center in North Carolina this spring. IBM is planning construction of the largest data center in China – nearly 6.7 million square feet. 6The Wall Street Journal, “IBM to Build Big Data Center in China,” by Owen Fletcher, 1/26/11
And in a veritable land rush in the Pacific Northwest, Google, Microsoft, Yahoo, and now Facebook have either built or plan to build data centers, incentivized by local tax breaks but more importantly, to be close to abundant supplies of cheap electricity, much of it supplied by the Columbia River. 7Wired Magazine, “The Information Factories,” by George Gilder, 10-06 Data centers are such power hogs that Google formed its own power company, Google Energy, to enable it to buy and sell power in the regulated wholesale market. 8Greenpeace International White Paper, “Make IT Green/Cloud Computing and its Contribution to Climate Change,” 3/10
Lastly, there's the money – lots of it. On the consumption side, businesses are looking to reap significant IT savings by moving data and processes to the cloud. On the delivery side, there's plenty of money to be made in satisfying that need. Amazon Web Services is expected to grow 50 percent in 2011, to $750 million, and is projected to have revenues exceeding $2 billion in 2014. RackSpace Hosting, another major player, had 2010 revenues of $100 million.9The Wall Street Journal, “Meet the Rainmakers,” by Nick Clayton, 2/15/11
In Europe, the Centre for Economics and Business Research projects that cloud computing will deliver more than €763 billion – that's more than US$1 trillion – in "cumulative economic benefit" (which includes IT savings) from 2010 to 2015, and directly or indirectly create more than 2.3 million net new jobs. 10Centre for Economics and Business Research, Ltd. Report for EMC, “The Cloud Dividend: Part One,” 12/10.
A ready and reasonably receptive market, rapidly developing infrastructure, impressive financials – the stage is set for the cloud to fundamentally change the way the world stores, manipulates, and accesses data.
Simple Concept, Complex Definitions
Outgoing Google CEO Eric Schmidt is credited with describing the vision, way back in 1993, that became a mantra for Sun Microsystems, where he was then CTO: "The network is the computer." 11Wired Magazine, “The Information Factories,” by George Gilder, 10-06 In its most basic form, this is the essence of the cloud – that the collective assets of processing power, storage, software and network are able to be harnessed anytime, anywhere, by users to whom it doesn't matter where anything is located.
The National Institute of Standards and Technology has arrived at this more complex official definition of cloud computing: "a model for enabling convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, servers, storage, applications, and services) that can be rapidly provisioned and released with minimal management effort or service provider interaction." 12http://www.nist.gov/itl/cloud/upload/cloud-def-v15.pdf
"It's a definition only a standards body could love," says Anshu Agarwal, vice president of marketing for Keynote Systems. "The way we think of it instead is that 'the Web is now the cloud.'With the open Web, open Internet standards and protocols, and the common denominator of http, it's advancing interconnectedness at a pace that is truly breathtaking."
In layman's terms, for consumers and business system end-users, the cloud means your digital stuff doesn't live in the box (or smartphone or tablet) that's in front of you. It's somewhere else. For enterprises, it means some or all of the corporate computing power and data has moved out of the corporate data center to somewhere else – and "somewhere" else can mean a variety of places (see sidebar).
The Four Types of Cloud
In meteorology, you have cumulus, cirrus, stratus, nimbus. In computing, you have private, public, hybrid, and community. Here's what each means as defined by NIST.
Private cloud. The cloud infrastructure is operated solely for an organization. It may be managed by the organization or a third party and may exist on-premise or off-premise.
Public cloud. The cloud infrastructure is made available to the general public or a large industry group and is owned by an organization selling cloud services.
Hybrid cloud. The cloud infrastructure is a composition of two or more clouds (private, community, or public) that remain unique entities but are bound together by standardized or proprietary technology that enables data and application portability (e.g., cloud bursting for load balancing between clouds).
Community cloud. The cloud infrastructure is shared by several organizations and supports a specific community that has shared concerns (e.g., mission, security requirements, policy, and compliance considerations). It may be managed by the organizations or a third party and may exist on premise or off premise.
Source: National Institute of Standards & Technology,http://www.nist.gov/itl/cloud/upload/cloud-def-v15.pdf
The Cloud's Silver Lining
Why is reaching for the clouds such a hot topic in the business world? Underneath it all is this irksome statistic: Some 70 to 80 percent of IT budgets are dedicated solely to keeping the lights on – maintaining hardware, software and network connections, and servicing users. The lion's share of IT budgets is sucked up by the status quo, not for adding functionality, streamlining operations, or gaining a competitive edge.
The cloud promises to change that and deliver impressive benefits in cost savings, agility, elasticity and simplicity.
Cost savings. The cost savings from cloud computing can come from a number of directions. The cloud puts the cap on CAPEX: The expense of buying and maintaining servers can be eliminated (which is why many organizations consider the move when their current hardware is near end-of-life). Software maintenance – patches, fixes, updates and upgrades – is owned by the cloud provider, reducing IT staff time and cost. And those significant energy expenditures noted above become the cloud provider's cost as well.
Agility. Freed from much of the maintenance burden, IT staffs can focus more on developing capabilities that respond to market changes, add efficiency to operations, and create competitive advantages. And they can appropriate cloud resources for their development work and testing, without provisioning in-house servers and bandwidth. The result is faster "time to value." Additionally, when the enterprise needs more computing horsepower, it's a simple, fast, self-service process through a Web portal to add it.
Elasticity. Retailers need tremendous capacity in November and December that would lie idle the rest of the year. Accountants hit their peak in March and April, with the other ten months quiet by comparison. "Traffic typically picks up in the end of Q3 and in Q4, typically four to six times higher than a normal week or month," says Alex Taranukha, senior director of architecture for the website ShopLocal.com. "Instead of putting in more boxes, we just add more service to our load balance from Rackspace."
And for CPU-intensive business processes, elasticity can mean having an army of machines available to quickly complete tasks that would take much longer on a company's own, smaller, internal array of machines.
Simplicity. Nothing goes out-of-date more quickly than computer hardware and software. Transferring the headache of keeping it all up-to-date to the cloud provider is a very attractive proposition. But it's not without trade-offs. "Under certain circumstances, the cloud-based services are often less feature-complete than the traditional [boxed] software," says Todd Kelley, VP of operations and managed support services for Competitive Computing (C2). "It's less configurable or customizable, and so some businesses are struggling with that trade-off – it costs less and is more scalable, but they can't always do everything their own special way."
It's Not All Blue Skies And White Clouds
That's a great list of potential benefits, but enterprises in general and CIOs in particular also have serious concerns about trusting their digital operations and data to third parties largely outside of their control. Data privacy and security is chief among them. Many companies have invested heavily in security infrastructure and procedures, and are understandably dubious about a cloud provider's ability (or willingness) to match their efforts. Co-mingling of data with that of other companies in co-tenancy scenarios is likewise a concern.
Reminders of weaknesses in the cloud appear routinely in the daily headlines. Recent case in point: Google loses all the data for 150,000 Gmail account users in the course of performing a storage software update. What CIO doesn't see that headline and picture his own users' reaction to such a catastrophe? And how does it get solved if it's all out of IT's reach and control?
Regulatory compliance is another area of concern for publicly traded companies and those in certain sensitive industries. Speaking about private-practice medical professional communities, C2's Kelley tells Benchmark, "We have options where they could have all [their records] off-premises in the cloud – options that cost less – but they don't really want much of anything to do with it."
Even assuming security and compliance issues can be satisfied, there's still a huge elephant in the room that threatens the success of cloud-based operations: performance.
Performance: The Cloud's Achilles Heel?
Think about all the steps in a hypothetical digital value chain. An order comes in through a branch office or store. Headquarters is queried for the customer account data. The manufacturer (third party) is queried on inventory and delivery dates. A request is sent to the logistics company to take delivery from the manufacturer and deliver to the customer. All the data goes back to corporate to prepare the invoice and update the customer records. Then the paperwork is sent back to the originating branch office or store.
In the old days, any of the transactions happening outside the branch office or corporate data center would involve phone calls and emails. Today, though, it all travels across the Web – the cloud – via both direct connections and Web APIs. The headquarters data center, and those of the manufacturer and logistics company, may or may not even be on their premises. They could be a virtual installation on the banks of the Columbia River.
How do you measure and monitor performance under such a scenario? If there's a problem, how do you pinpoint it along this convoluted chain?
If it was simply a website, field agents could be used to measure performance to the end user anytime, anywhere, as Keynote has done for a decade and a half. If it all happened inside the corporate data center and within the company's private WAN, measurement agents could be installed at appropriate locations. But today's digital value chain is both of these and more.
"In the old enterprise world, you would install either hardware or software agents, 'listening posts,' throughout your architecture, throughout your infrastructure," Keynote's Agarwal says. "But you don't own the entire environment now. And your partner isn't allowing you to put little listening agents in their infrastructure, so you really don't have visibility anymore. The only time you know something is going wrong is when applications are failing for users. And how do you possibly find out what's gone wrong and where if you don't have visibility into the whole transaction sequence?"
Whether it's a private, public, or hybrid cloud, IT still is held responsible for performance. And so IT has to find a way to see how each of those links in their digital value chain is performing, even when it involves the black boxes of cloud providers and/or inaccessible partner/supplier data centers. That means having agents monitoring every link in the chain – at point of termination, inside and outside the corporate firewall, and at the first and last mile. It's more complicated, but achievable – and imperative.
"Our public infrastructure, our global monitoring network is very well-positioned to monitor applications whether they're hosted in a public cloud or in a private cloud," says Keynote's Ian Withrow, senior product manager. "Now we're able to go inside the firewall and monitor all the point-to-point links and traffic within a network."
The popular retail deal-aggregator ShopLocal.com has this type of complex architecture and has put in place a similarly sophisticated monitoring system deployed by Keynote.
"Using Keynote’s portable private software agents helped us a lot to monitor first-mile performance and API endpoints. And at the same time, Keynote’s public monitoring network let us monitor Akamai and the last-mile performance the end-user sees," says ShopLocal’s Taranukha. "So it actually gave us very good insight into how different parts of the architecture perform, and helped us to allocate our hardware and resources at Rackspace, our hosting company, very efficiently."
The Performance Price Tag
Minute by minute, degraded performance exacts a painful toll when mission-critical operations are involved. Research of IT executives conducted by consulting firm Enterprise Management Associates indicates the cost of downtime "clusters" around $60,000 an hour. Some 40 percent of those surveyed reported 30 minutes or more of unscheduled downtime in any given month. 13Enterprise Management Associates, “Why Web Performance Monitoring Is More Than A Competitive Advantage for SaaS Providers,” 7/10
"For businesses with critical apps, our average estimate is $100,000 a minute," says EMA Vice President Dennis Drogseth, adding, "I was talking to a group about downtime for financial transactions. It was a million dollars a minute. So you probably wouldn't fully depend on an external cloud provider for that app"
SLAs as Elusive as a Cloud
One reason cloud performance monitoring is so critical is that cloud provider service is so nebulous. "Performance" to cloud providers typically means only availability; and even availability is only loosely guaranteed. For Amazon Web Services, as an example, unavailability means no connectivity at all during a five-minute period; if your user has a lousy, erratic, miserably slow connection, as far as Amazon is concerned, they've delivered. And availability means availability when it leaves Amazon's door; whether or not anything actually reaches your user is not Amazon's problem (regardless of their choices for ISP and connectivity). Oh, and the burden of proof is on you, the customer. For all intents and purposes, Amazon is not even checking to see if you even have service.
This is not to dump solely on Amazon. The same guarantees, or lack thereof, are typical of many cloud providers. In addition to the caveats above, scheduled and emergency downtime is excluded from availability guarantees; penalties for unavailability are minimal, and certainly not commensurate with potential business damages; and any other kind of performance is not included in the service level agreement.
"Many cloud service providers don't offer anything beyond junk for SLAs," Drogseth says, (see accompanying interview). But, he continues, "Service level is key. If you're an enterprise and your cloud provider is not committing to service levels that you feel comfortable with, kick them out. The industry will change to a commodity set of service providers who don't care about you. A more partner-oriented group that will start to give you visibility as you need it – that's the group you want to work with."
An ideal cloud-client working relationship includes substantial SLAs, external monitoring of SLA parameters that is visible to both provider and client, and meaningful recourse if the service falls short. In lieu of this ideal, however, the onus is on the enterprise to put monitoring and measurement in place and to hold their provider accountable – so they can either get the service level they need, or switch to a better provider.
End-users don't necessarily care where or how their applications are delivered; they just care that they work. And the people they hold responsible are the IT staff.
"In the end," Drogseth says, "your products or applications as an IT organization – whether you harvest them through providers or whether you develop them, whether you deliver them partly through the cloud or not –will stand or fall on the success of the quality of your application services and your ability to deliver them cost-effectively and responsively."
Given the "cloud" moniker, meteorological analogies are appropriate (if overused, even here), and the forecasts are similarly best validated by a look out the window. But given the compelling case for offloading computing to the cloud, and the tremendous investment happening in cloud infrastructure, it's safe to say that cloud computing is a snowballing phenomenon that is going to take over a significant part of the digital world.
The network is becoming the computer. It's up to enterprise customers to determine, through rigorous performance monitoring and measurement, whether it will be a super-fast, user-pleasing, productivity-boosting workhorse, or a mediocre-performing, user-annoying, productivity drag.