Cloud computing … what the heck is it? Not only do I not possess the magic definition (and I talk about it nearly every day), but I’m not convinced anyone does.
Structure 08, held last week in San Francisco, offered attendees the opportunity to wrap their heads around the idea of cloud computing, bringing in some of the biggest names and companies in the industry — Amazon, Google, Microsoft, GigaSpaces, Salesforce.com, etc. — to try to clear up the picture. But therein lies the rub: everyone does “cloud” differently, and depending on your semantical leaning, anyone can be found to offer more true-to-definition cloud computing than the other guy. Is cloud computing open? Should it be? Well, that depends on your definition of “open.” You get the point.
Actually, this very question (of cloud openness) stole the show in a panel discussion featuring: Christophe Bisciglia, senior software engineer with Google; Jason Hoffman, founder and CTO of Joyent; Tony Lucas, CEO of XCalibre Communications; Lew Moorman, senior vice president of strategy and corporate development at Rackspace; Geva Perry, chief marketing officer of GigaSpaces; and Joe Weinman, vice president of strategic solutions at AT&T. But the discussion didn’t limit its mixed definitions to the realm of openness.
The discussion kicked off with a question by moderator, and BitCurrent analyst, Alistair Croll asking the panelists about the differences between cloud computing and grid computing. Joyent’s Hoffman answered that grid computing is about batch jobs, I guess inferring that cloud computing is inherently more on-demand. While the latter might be true, the distinction can’t be that clear. After all, banks are increasingly using grids for algorithmic trading — a practice that cannot rationally be described as batch processing. AT&T’s Weinman, for his part, drew the line at billing. A cloud that isn’t billable, he said, is a grid. Again, it cannot be — and, indeed, is not — this easy.
One of the biggest arguments in cloud computing is whether it must involve an external service, or if internal platforms can be considered clouds. If the latter is true, as companies like GigaSpaces, Appistry and even IBM believe to be the case, then the difference between clouds and grids cannot be broken down to billing, or to a case of private versus public. Blurring the lines even more is the case of a company using a grid infrastructure with a billing mechanism to charge the individual departments within the company. What do we call it then? It’s internal, it’s a grid, but it’s billable, so it must be a … cloud?
No one with whom I spoke at Structure (or elsewhere, for that matter) seems to have a definitive answer to this question. I posit that there is no blatant distinction between grid computing and cloud computing because the latter represents the service-oriented evolution of the former. Especially given the wont of marketing departments to stretch the definitions of buzzwords to fit whatever they’re doing, I predict we can expect to see almost every service-oriented distributed platform described as a cloud — even if we would have called it a grid just a year ago at this time. My advice: deal with it.
As noted above, the highlight of this panel was a lively, at times heated, debate about whether cloud computing is (or should be) open, or whether it is subjecting users to the type of lock-in we have decried in traditional models. XCalibre’s Lucas and Joyent’s Hoffman called out Google, claiming that its use of Big Table as the database layer of App Engine forces users to abandon existing databases and write only with Big Table in mind. Google’s Bisciglia defended these allegations deftly, stating that the App Engine API works best with Big Table, but doesn’t require developers to use it. He added that there are competitive reasons for not completely opening up the Big Table code.
Let’s be frank, though. Google most likely created App Engine with the thought that users would leverage Big Table; openness probably wasn’t a big concern. Is this such a problem? As Bisciglia put it, Big Table is something that has worked very well for Google, so it just logical Google would make it the database component of any cloud computing platform. The bottom line is that it is the best choice right now. This sentiment was echoed by Rackspace’s Moorman, who noted that cloud computing sometimes is about trade-offs, wherein customers wanting the best tools might have to make some sacrifices. It’s not necessarily that Big Table is competing with Oracle, but rather that Big Table is just a part of the App Engine platform. Mosso, the cloud computing division of Rackspace, offers a similar platform, where, in the name of optimal performance, users are relegated to porting only applications written in languages that best mesh with the Hosting Cloud platform.
While everyone agreed that cloud standards and interoperability are noble goals that would solve many of the problems relating to cloud provider lock-in, I don’t necessarily see them as too important just yet. Just like everyone seems to have different definitions of cloud computing, so do people have different ideas of what it will take to get enterprises on board. I tend to agree with Moorman in believing that trade-offs are not necessarily such a bad thing, and that the optimized tools (and, thus, limited flexibility) of highly managed offerings like Google’s App Engine or Mosso’s Hosting Cloud actually make life easier for newcomers. Once they experience positive results with minimal effort, maybe they will feel comfortable diving a little deeper, taking advantage of solutions like Amazon EC2 or GoGrid, near-bare-metal offerings which allow greater flexibility in terms of programming language, operating system, database, etc.
Other Structure Highlights
Jonathan Yarmis, vice president of disruptive technologies for AMR Research, described cloud computing as a real paradigm shift, insinuating that this is not a “boy who cried wolf” situation. The next five years could make the Internet revolution “feel like child’s play,” he said, with cloud computing and stream computing enabling EaaS — everything as a service. He added that with the explosive growth in data, especially of the disposable variety variety, cloud computing will be called upon to manage and process it. Yarmis concluded by answering an audience question about how the economic downturn could affect cloud computing. With its higher efficiency and cheaper entry points, “I think an economic downturn, in fact, could accelerate cloud computing,” said Yarmis.
Amazon.com CTO Werner Vogels led off his keynote by stating that, “At Structure 10, the whole discussion will be different. All the things that now seem new will be established.” If Amazon’s history is any indicator, Vogels’ prediction will be spot on. He noted how Amazon’s infrastructure was a “duct tape and WD-4- operation” until 2001, when a project to optimize Target’s retail Web site helped transition Amazon’s platform into the service-oriented (currently housing about 1,000 services) we see today. Last quarter, he said, Amazon Web services bandwidth flew past the company’s e-commerce bandwidth, adding credence to a ZDnet quip that with its Web services, Amazon is like a bookstore selling cocaine out the back door. Guess which one is the real business. (Other points of Vogels’ keynote were the same as his Next Generation Data Center keynote last year.)
Sun Microsystems CTO Greg Papadopoulos took advantage of his afternoon keynote to plug Sun’s Project Caroline, which is described on the project site as: “a research program developing a horizontally scalable platform for the development and deployment of Internet services. The platform comprises a programmatically configurable pool of virtualized compute, storage, and networking resources. Project Caroline helps software providers develop services rapidly, update in-production services frequently, and automatically flex their use of platform resources to match changing runtime demands.”
Papadopoulos also predicted that by 2010 (curiously, the same year noted earlier by Vogels), Web, HPC and SaaS workloads will comprise the majority of system volume.
Mendel Rosenblum, co-founder and chief scientist of VMware, provided some of most insightful nuggets of the day, particularly as they relate to VMware’s vision. In an on-stage interview, Rosenblum said that server consolidation was just an entry point, but the real vision of VMware always was about decoupling all computing from specific hardware and experiencing dynamic allocation via a distributed system running your entire datacenter. You can basically run your own cloud now, he said, and (thanks in part to multi-core processors) it “doesn’t even make sense” to run a machine without a virtual layer.
Speaking of cloud computing, Rosenblum says people are free to have “cloud visions” where no one is managing machines, but VMware sees users being a little more conservative. He sees companies building their own little clouds, maybe migrating if they get comfortable with large-scale providers. And if they do decide to make the move, Rosenblum said that by the end of next year, VMware should have the technology to allow for seamless switching between in- house and external clouds. Adoption, however, will be more gradual, he believes.
- Ot notable absence from the stage was IBM, who some would say is leading the enterprise cloud computing charge with its soon-to-come Blue Cloud line of solutions.
Final Structure Takeaways
Although Structure failed to give attendees a clearer understanding of what, exactly, cloud computing is and is not (if that even was one of its goals), the event certainly was full of excitement and full of optimism. If Structure continues in the years to come, and if more events spring up with the goal of highlighting what we are today calling “cloud computing,” I have to think a definition will firm up right along with it. Of course, as long as users are excited and actually adopting these technologies, whatever we call them, a definition might not matter. And as I heard analogized by several speakers and attendees, deciding to adopt cloud computing is a lot like deciding where to put your money: people initially were skeptical about putting their money in the bank until they realized that the bank was a better option than sticking it under the mattress. As cloud services and their respective providers continue to evolve, the IT community will make the same determination.
Elsewhere in the Issue
For an in-depth look at Hyperic’s new tool for monitoring the status (availability, throughput, latency, etc.) of Amazon’s stable of Web services, be sure to check out Dennis Barker’s feature article “Shining a Light into Amazon’s Darkness.” Many are skeptical about what is happening within EC2 in terms of performance, etc., and Hyperic’s CloudStatus.com is among the best tools available for looking into this cloudy area. In time, Hyperic plans to offer similar capabilities for Google’s App Engine and Salesforce.com’s Force.com platforms, and also hopes to incorporate application-specific performance metrics.
There also were many big announcements across the cloud computing and broader virtualization market. For the sake of brevity, I’m only noting the following, but there are several others of some import, so be sure scan the entire issue: “Yahoo Realigns, Focuses on Cloud Computing”; “IBM Opens Cloud Computing Centers in Africa, China”; “Univa UD Unveils New Datacenter Automation Strategy”; “Microsoft Deploys IBM iDataPlex in HPC Labs”; “GigaSpaces Offers Scale-Out App Server on EC2”; “Symplified Intros Grid-Based Identity Management for SaaS”; and “DynamicOps Intros Unique VM Management Software.”
Comments about GRIDtoday are welcomed and encouraged. Write to me, Derrick Harris, at [email protected].