Cloud Cover

Much has already been written about Chrome, a new open source Web browser that Google announced today. I am not going to rehash the basic facts about Google’s move. While Chrome’s re-imagining of the browser as an application substrate certainly excites me as a geek, this is really the first attempt by a major cloud computing vendor to provide a front-end to its cloud.

Of all things about Chrome, I am most surprised that it took so long to be released and that it did not build on one of Google’s existing footholds on client computers, Google Desktop. Amid all the furor about this release re-igniting browser wars of a bygone era, it is easy to miss the point that with this product, Google is marrying a Web browser and RIA runtime to advance on what is surely the weakest link in its dream of organizing the world’s information and making it accessible through the Google Grid, the traditional document-rendering Web browser. Details on the code behind the Google Chrome project are scant, but it’s clear Google intends to spur innovation in Web browsers by atomizing them into their main component parts: a rendering engine, a plugin architecture and an honest-to-goodness virtual machine for Javascript. In this way, Chrome is not just a re-imagining of the user experience of a Web browser but also how a browser is developed.

In a way, it is only appropriate that Google Chrome be the front end to Google’s cloud infrastructure. The back end of a typical Web application has morphed into a flexible cloud with the addition of virtualization, a layer of abstraction. Likewise, Google Chrome provides a user interface meta-frame for combining and recombining Web applications flexibly via tabs and browser plugins.

Despite all the above points, Google Chrome comes into a market with two formidable incumbents, Internet Explorer and Firefox. As cool as Chrome sounds, it needs to prove itself to be much better than existing browsers and RIA runtimes for users of existing browsers to even bother making the switch.

What would make you seriously consider using Chrome as your primary browser? Please join the discussion in the comments!

Cassatt’s Ken Oestreich gives a shout-out to our forthcoming power management report and webinar. When I was working out of our London office I had the chance to spend some time with Andy Lawrence, our eco-efficient IT analyst and the primary author of the report. His mighty brain is awesome! As with our open source practice (and, indeed, this infrastructure practice) I am excited about the potential of Andy’s work to help people get things done in a correct and forward-looking way. Eco-efficiency isn’t just a nice-to-have for CSER brownie points any more. It’s good engineering and smart accounting. Those are two of our favourite things.

Whether internal compute clouds are real clouds or not, one thing people would really like is the ability to move computing jobs between internal IT resources and a service provider partner. We’ve taken to calling this cloudbursting - dealing with traffic bursts by dynamically obtaining excess capacity from the cloud. It turns out Target’s been doing this for years, coping with its Christmas peak loads by offloading onto Amazon Web services.

The trick, of course, is having a nicely-defined blob of computing stuff you can push up into the cloud, or pull back down to the IT shop. And if there’s one thing virtualization’s good at, it’s neatly encapsulating blobs of compute stuff. That’s where CohesiveFT comes in. Originally a vendor of special-purpose virtual appliances for the financial industry, it’s now working in application-deployment-via-virtualization. This week Cohesive announced support for Amazon EC2 large, extra-large and high-CPU instances. This should help the classic AWS user - a social networking app in San Francisco’s SoMa district, for example - ramp up very rapidly into production.

The other thing people would like to be able to do is compare clouds, at least with each other and ideally with internal resources as well, on metrics like performance and cost. We think the hero and heroine CIOs of the future will be those best equipped to decide where those compute blobs should run. Hyperic took a big step forward with its CloudStatus.com, originally a monitor for AWS. Now CloudStatus monitors Google App Engine as well. Take Hyperic up on the offer to integrate these services with your inhouse Hyperic HQ deployment (at a price), and you’re well on the way to having your head in the clouds (and your feet firmly planted on the ground.)

Ken Oestreich from Cassatt writes a timely post to address the oft-encountered notion that cloud computing must come from providers external to the enterprise. He goes on to explore how a cloud computing infrastructure may be created from existing (’legacy’?) IT investments and some ways in which such an architecture would be used. Ken references some misgivings industry folks have about moving to a cloud computing infrastructure, and I honestly don’t blame them. Although enabling technologies and economic models for cloud computing have been around for some time, the biggest hurdle to overcome such a big switch is in fact a cognitive one.

The 451 Group’s thinking that we have set forth in our position paper about cloud computing is that enterprises will probably end up adopting a standard-based cloud computing architecture that relies on internal and external infrastructure. One little quibble I have with Ken’s post is that it equates cloud computing with utility computing. Now I know I am well into buzzwordland already when expounding on cloud computing, but we think it is much broader than utility computing in that it is both a technological and economic organizing principle. Roughly speaking–and I emphasize ‘roughly’ here–we think cloud = utility + strong retail discipline.

Although the relevant enabling technologies have been around for some time, retooling existing assets into a cloud infrastructure isn’t easy and doesn’t come with strong enough incentives to use it. Merely building a fine-grained internal chargeback infrastructure in an enterprise doesn’t mean it will be used. When enterprises are considering a switch as big as the one to a shared cloud architecture, starting out with sourcing compute capacity from external providers forces a retail discipline at the edges of every transaction, monetary or computational. Once teams have learned to work with this discipline, it can be extended to internal assets that have been brought into the fold of the cloud architecture. I don’t mean to knock Ken’s larger point though–if anything, by anticipating the market need for incorporation of existing infrastructure into an internal cloud, companies like Cassatt, Elastra and Enomalism are clearly on the ball.

Internal clouds undoubtedly have their place but don’t necessarily sit atop accurately provisioned infrastructure. Public cloud computing is what enables Joe IT Guy (or even Karen Business Analyst) to pull out a credit card and add in extra capacity during times of peak load. We think that once there are enough public cloud providers with adequate  SLAs, most of the points that give enterprises the heebie-jeebies will be allayed (notable exceptions being the ones that are created by external regulation/legal requirements). Like Ken, we also think the concepts of cloud computing will eventually percolate into the datacenter, but that will come to pass only after externally-sourced cloud compute capacity has put enterprises through a certain rigorous discipline and transformed the thinking of IT managers.

Do you think retooling internal infrastructure into a cloud would be enough of an incentive to adopt cloudy thinking? Please join the discussion below in the comments.

First there was the Gmail outage:

Many of you had trouble accessing Gmail for a couple of hours this afternoon, and we’re really sorry.

Then came a major problem with VMware ESX/ESXi 3.5 Update 2:

An issue has been discovered by many VMware customers and partners with ESX/ESXi 3.5 Update 2 where Virtual Machines fail to power on or VMotion successfully.

In each case the company scrambled to do the right thing by its customers.

But following as they did on the heels of the July 20 Amazon S3 “availability event”, these incidents serve as sobering reminders of the costs of over-reliance on a single, fallible platform vendor, regardless of whether that platform is software or service. No wonder people are starting to talk about Redundant Arrays of Inexpensive Clouds. Virtualization and the cloud change the nature of computing, sure, but they don’t transcend all human frailty. Keep backups, people, and have contingency plans. Let’s be careful out there.

Cloud configuration startup Elastra has raised $12m in its second round of funding led by Bay Partners with new participation from Amazon.com and existing investors Hummer Winblad Venture Partners. Elastra enables enterprises to begin deriving some of the benefits of cloud computing by letting them push legacy systems and existing IT investments into a cloud. Elastra competes with startups like Enomaly, 3Tera, RightScale and Kaavo.

I meant to blog an item I saw earlier this week about Dell seeking to trademark the term Cloud Computing. According to an excerpt from the filing quoted at the above link, Dell wants this filing to cover not just a proprietary cloud computing platform but general use of the term cloud computing, which it defines as

IC 040. US 100 103 106. G & S: Custom manufacture of computer hardware for use in data centers and mega-scale computing environments for others

IC 042. US 100 101. G & S: Design of computer hardware for use in data centers and mega-scale computing environments for others; customization of computer hardware for use in data centers and mega-scale computing environments for others; design and development of networks for use in data centers and mega-scale computing environments for others; Consulting services for data centers and mega-scale computing environments in the fields of design, selection, implementation, customization and use of computer hardware and software systems for others; Consulting services for data centers and mega-scale computing environments in the fields of design, selection, implementation, customization and use of computer hardware and software systems for others

What initially raised my hackles is that Dell’s definition of cloud computing is a bit too focused on hardware for our taste. Although appropriate hardware design for cloud computing datacenters is important, it is only one of the many technologies, including virtualization, that enable cloud computing. We would add in the economic model of cloud computing, pay-as-you-go only for what you use–as another key characteristic. What damns Dell’s definition, in my mind, is the inclusion of consulting services for datacenter design and customization. One of the big promises of cloud computing is low barriers to entry and the relative absence of high-touch engagements such as consulting contracts. As my esteemed colleague James Governor of Redmonk puts it, “If there is a consultant in the room… its not a cloud.”

To be sure, cloud computing as a general organizational principle for computing is enabled by a well-defined set of enabling technologies and economic models. But one thing I particularly like about cloud computing is its vagueness-by-design, which enables us to think big about the possibilities enabled by this new model of computing without necessarily getting trapped in semantic rabbit holes. There is speculation that Dell’s attempt to trademark this term is a defensive move related to its ownership of cloudcomputing.com. Regardless, trying to put too fine a point on the general definition of cloud computing via a trademark may be a futile exercise that doesn’t necessarily advance the dialogue on this promising new model of computing and may even serve to hold it back.

Have you ever noticed that you can see many more stars in the night sky through your peripheral vision rather than by focusing directly on patches of the sky? Likewise, if we keep cloud computing in your peripheral vision but not in your direct sight, we may be better able to understand what we can accomplish with it.

My big kid went through the San Francisco public school lottery this year. That was an experience. LA and New York have school lotteries as well, so parents there will know the drill. I toured a dozen schools, put seven choices in order, entered the lottery in January and got her school assignment in March. Our family was one of the lucky ones. The kid was placed in a great neighborhood school where she will be in a Spanish immersion class. Stay with me; this is about infrastructure computing.

The San Francisco school board gets a lot of flack from parents, much of it deserved, but its Spanish immersion programs are renowned. As such they’re in hot demand, not only from Type-A Anglophone moms like me, but from families where Spanish is spoken at home. Lots of those families want their children to keep their mother tongue while acquiring English as well.

The immersion programs are supposed to enrol 50% Spanish-dominant and 50% English-dominant children in each class. To the distress of some of the Spanish-dominant families, the classes at two of the schools enrolled overwhelmingly English-dominant children. It turned out that English siblings had been mistakenly coded as Spanish. (Younger siblings automatically get placed at the same school as their older brothers and sisters.)

The lack of Spanish-dominant role models in the classroom meant that the children would not learn Spanish as well as if there were a 50/50 ratio between the languages. It also meant that the Spanish families had been unfairly shut out. The school district eventually decided to reassign some English-dominant families to other schools. This was announced four weeks before school is scheduled to start, and has caused considerable distress and hardship to the families involved.

And (here’s my point, you knew it was coming) it was all for want of a horseshoe nail; or in this case, decent infrastructure computing for the school board’s Educational Placement Center (EPC). Not only were the mis-identified children hand-coded into the system (we couldn’t automate that?); the database used by the EPC keeps crashing, and every time it crashes, data has to be re-entered by hand.

DUDE. Are we in the past?

The stats are here. About 12,000 families entering three grade levels (Kinder, 6th grade and 9th grade); each family listing up to seven choices in order; optimize so each family gets the highest possible choice. It’s a big complex job that only needs to be run two or three times a year. Cloud or grid computing providers - you want to earn the undying love of San Francisco parents? And get CSER brownie points by solving a hard problem for an underfunded school district? The case study practically writes itself.

Me and my trigger finger just deleted some more comments. I am sorry. If you commented and it hasn’t appeared, it’s nothing personal: the problem is between *my* keyboard and chair.

The aptly-named Hugh Macleod, maker of the Gaping Void cartoons, makes a rather chilling point about clouds:

…nobody seems to be talking about Power Laws. Nobody’s saying that one day a single company may possibly emerge to dominate The Cloud, the way Google came to dominate Search, the way Microsoft came to dominate Software.

“Monopoly issues aside, could you imagine such a company? We wouldn’t be talking about a multi-billion dollar business like today’s Microsoft or Google. We’re talking about something that could feasibly dwarf them. We’re potentially talking about a multi-trillion dollar company. Possibly the largest company to have ever existed.

This underscores the importance of the Franklin Street Statement:

Service providers are encouraged to…

• Make data and works of authorship available to their service’s users under legal terms and in formats that enable the users to move and use their data outside of the service.

Without the power to grab our data and walk away, we are at the mercy of our providers. And unless we have that power, they have little incentive to be merciful.