2008年10月17日金曜日

Cloud Computingに対する辛口の考察

Cloud Computingを一つのトレンドとして捉えて、最後には何も残らないのでは、という記事も多く現れてきている。  本記事は、こういった記事を全体的に捉えて、何が残りうるのか、を分析している。 記事によると、次の技術は生存しニッチ市場を確保するだろう、との事。
  1. "Cloudbursting" -
    企業内のアプリケーションニーズで短期間のアプリケーション開発、短期間トラフィックが急増するアプリケーションに対応するサービス

  2. Cloud OS -
    データセンタ全体を一つのOSとして、CPU、メモリー、ネットワーク、ディスク、他を統括できる環境
  3. SaaS/PaaS -
    これは現実にこのモデルで成功しているベンダーがいっぱい増えている、という現実があるから。
  4. Mashups/WOA -
    Mashupと呼ばれる、いろんなWebアプリケーションを組み合わせて、新しいユースケースを産むWeb アプリケーション市場。 
  5. "Quick start" data centers and data center modules
    GoGridやFlexiscale、Amazon Web Serviceのように、パッと始めて、必要なくなったらパッとやめれるようなデータセンタ環境
いづれも日本市場では成立しうる市場ではないか、と考える。 
 

Cracks in the Clouds, but the Sky Ain't Fallin'

Update: I accidentally left off the reference links in the first paragraph. This is corrected now. My apologies to all that were inconvenienced.

This last couple of weeks have been filled with challenges to those preaching the gospel of cloud computing. First it was a paper delivered by three Microsoft researchers describing in detail the advantages of small, geo-diverse, distributed data center designs over "mega-datacenters", a true blow to the strategy of many a cloud provider and--frankly--large enterprise. Second, the Wall Street Journal published a direct indictment of the term, cloud computing, in which Ben Worthen carefully explains how the term ended up well beyond the boundaries of meaning. Added to the dog pile was Larry Elison's apparently delightful rant about the meaninglessness of the term, and an apparent quote where he doubts the business model of providing capacity at scale for less than a customer could do it on their own.

Frankly, I think there's some truth to the notion that cloud computing and many of the notions that people have about it are beginning to lose their luster. We seem to have passed through a tollgate of late, from the honeymoon era of "cloud computing will save the world" to the evolutionary phase of "oh, crud, we now have to make this stuff work". While the marketing continues unabated, there are some stories creeping out of "cloud-o-sphere" of realizations about the economics and technical realities of dynamically offloading compute capacity. Solutions are being explored to "the little things" that support the big picture: monitoring, management (both system and people) and provisioning. Gaps are being identified, and business models are being criticized. We are all coming to the conclusion that there is a heck of a lot of work left to be done here.

Doubt me? Take a look at the following examples of "oh, crud" moments of the past few months:

  • I can't for the life of me find the link, but about three months ago I read a quote from one of the recent successful Amazon EC2-based start ups noting that as their traffic and user base grows, they believe the economics of using the cloud will change, and moving some core capacity to a private cloud might make more sense.

    Update: John M Willis pointed me to the reference; a quote from item 8 of his "10 Reasons for NOT Using a Cloud" post, which in turn references a Cloud Cafe podcast in which "Brad Jefferson the CEO of Animoto suggested at some point he might actually flip the cloud." Read the post and listen to the podcast for more. Thanks, John.

  • Mediafed's Alan Williamson presents a keynote at CloudCamp London in July in which he notes that "[w]e've come to realize we cannot rely on putting all our eggs in one basket", and shows off their dual-provider architecture utilizing Amazon EC2 and Flexiscale.

  • A court case in the United States demonstrates the legal perils that still have to be navigated in terms of constitutional protections and legal rights for those that place data in the cloud. This case goes to prove that too much depends on the Terms of Service of the providers today to provide a consistent basis for placing sensitive data in the cloud. Even mighty Amazon cannot be trusted to run a business infrastructure alone.

Some are even hinting that cloud computing is stupid, and that it will fail to be the disruptive technology it is touted as being.

That last statement is where I part ways with the critics. Cloud computing--all of it, public and private--will be disruptive to the way IT departments acquire and allocate compute functionality and capacity. To me, this statement is true whether or not it turns out that it would be better to build 500 small, manageable, container based data centers than 5 megaliths. It will be true even if the term gets used to describe anti-virus software. There is great momentum pushing us towards huge gains in IT efficiency, and it makes little economic sense not to follow through on that. Like any complex system, there will be winners and losers, but the winners will strengthen the system overall.

Here's where I see winning technologies developing in the cloud:
  • "Cloudbursting" - This is the most logical initial use of the cloud by most enterprises; grabbing spare capacity on a short term basis when owned capacity is maxed out. It virtually eliminates the pressure to predict peak load accurately, and gives enterprises a "buffer zone" should they need to scale up resources for an application.

  • Cloud OS - The data center is becoming a unit of computing in and of itself, and as such, it needs an OS. However, the ultimate vision for such an OS is to grow beyond the borders of a single data center, or even a single organization, and allow automated, dynamic resource location and allocation across the globe from an open market system. That's the goal, anyway...

  • SaaS/PaaS - Most of my readers will know the SaaS debate inside and out: is it better to take advantage of the agile and economic nature of online applications, or is it both safer and, perhaps, cheaper in the long term to keep things in house? I think SaaS is winning converts every day and will likely win nearly everyone for some applications. PaaS gives you the same quick/cheap start up economics as SaaS, but for software development and deployment infrastructure. I'll post more on PaaS soon.

  • Mashups/WOA - Much has been said of late about the successes of loosely coupled REST-style Internet integrations using published URL-based APIs over the traditional "contract heavy" SOAP/WS-* world. It makes sense. Most applications don't need RMI contracts if all they are trying to do is retrieve data to recombine with other into new forms. If it remains as easy as it has been for the last five years or so, mashups will be an expected component of most web apps, not an exceptional one.

  • "Quick start" data centers and data center modules - Between private clouds made of fail-in-place data centers in shipping containers, and powerful Infrastructure as a Service offerings from the likes of GoGrid, Flexiscale, Amazon and others, both startups and large enterprises have new ways to quickly acquire, scale up and optimize IT capacity. Acquiring that capacity by traditional ways is starting to look inefficient (even though I have seen no proof that this is so, as of yet).

Even if it never makes sense for a single Fortune 500 company to shut down all of their data centers, there will be a permanent change to the way IT operations are run--a change focused at optimizing the use of hardware to meet increasing service demands. Accounting for IT will change forever, as OpEx becomes dominant over CapEx, and flexibility is the name of the game. Capacity planning is changed forever, as developers can grab capacity from the corporate pool, watch the system utilization as demand grows, tuning the application as needed, adding hardware only when justified by trend analysis. Start up economics are changed forever, as building new applications that require large amounts of infrastructure no longer requires infrastructure investment.

CloudCamp SV demonstrated to me that the intellectual investment in cloud computing far surpasses mere marketing, and includes real technologies, architectures and business models that will keep up on our toes for the next few years.