Recently in Cloud Computing Category

Much has been written about the severe Amazon EC2 outage last week. This made me think about the necessary tools needed for deploying high-availability applications in a cloud environment. Java Enterprise Edition is very complex to use but remains the popular choice among enterprise application developers today and it has a huge installed base needing some form cloud readiness. Application platform frameworks, like Spring, provide the runtime middleware container for both custom or packaged applications that run on a cloud service like PaaS (Platform-as-a-Service). Features of programming languages like Java, C#, C++, Ruby or Python; can be extended at runtime by APIs, embedded declarative clauses or metadata patterns provided by a framework like Spring. Optimal allocation of system resources (memory, threads, connection pools etc.), quality-of-service (reliability, availability etc.) and connectivity (messaging, networks & databases) are managed on behalf of the application by the framework. With the huge investment in Java code today, many firms are adopting Spring's Model-View-Controller (MVC) for web applications, plug-ins for the Eclipse IDE and the many web service add-ons available. The beauty of the framework and its relevance to cloud is that Spring separates all business logic from application (logical or physical) infrastructure. Now we can have a real application bus where you combine virtualized or non-virtualized applications with structured or unstructured data. Those applications will be exposed to both managed and unmanaged mobile devices (but that's another blog post!). We would build applications by taking blocks (objects) of code using Spring and populate the business logic using open source lifecycle tools existing outside of the cloud. Dynamic programming models like Ruby can be used as large-scale web-based front-ends with the power of Java EE under the hood extending the life of existing legacy applications in a painless way. With the framework in place, cloud advances in multi-tenant governance; horizontal scaling or cloud transaction processing can take place without causing major application reconstruction.

The news surrounding Intel's recent announcement of the Atom Z6xx (aka Moorestown) System-on-a-Chip (SoC) tends to focus on the uphill battle the company is facing in the ARM-centric smartphone ecosystem. Intel has claimed idle times of 21-23 milliwatts for the Z6xx series compared to 25 mW for a 1GHz Snapdragon. That is 10 days of standby time with a 1500 mAh battery. What is more interesting is the move to port Windows Server to multi-core ARM processors manufactured at 40 nm, such as announced by Marvell Technology Group. The chips will bring more than a five-fold reduction in power consumption in data centers and cloud environments compared to the x86. Think of the headroom an ARM implementation in servers would be when comparing an Intel Xeon at several hundred dollars versus a $35 ARM quad-core running virtualized Windows Server 2008. Om Malik brings up in a recent post that it was "too bad Intel sold its StrongARM technology to Marvell." I agree; Marvell did what Intel didn't have the heart to do. We think of virtualization in data centers as smart economics in hardware utilization and power consumption but what happens when server hardware processor cores decrease by a factor of 10 and power consumption by 5? Do we throw hardware at the problem again? Analysts should model the financial scenarios factoring in VMWare licensing costs, power consumption/footprint of rack space and application-specific-servers vs. general purpose power-hog blades running VMWare.

Beyond the hype around cloud computing and its variants, there is growing acceptance of the notion that the buying criteria of end customers has moved from a discussion of how the technology chosen and provided to a discussion of what value the technology is providing. This provides the new frame of reference for customers and IT providers alike.  We are finally getting to fully connect the business objectives and requirements with IT services delivery. Given IBM's announcement today of a new data center in Auckland and a Cloud Computing application research center in Hong Kong, there is growing evidence that organizations are increasingly more interested in moving much of their data center operations to off-premise providers. When we set aside details such as massively scalable processing, virtualization, service orientation and always-on access; we have nothing short of a major evolution of business itself. The key will be in pricing strategies of such services because smart IT shops can always reverse engineer your cost structure for standard cloud platform offerings as they consider the option of "private clouds" (a term I utterly detest). IBM references "private clouds" as those configurations where customers have dedicated computing and resources for their own business. They are compute and infrastructure technology stacks, nothing more magical than that. The provider profit margin and value will result from their ability to optimize everything behind the service boundary while hiding complexity from the end customer. That is something worth paying for!

The Apache Hadoop Core is an open source platform enabling developers to write and run applications across clusters of commodity computers and process vast amounts of data. As a primary investor and developer of Hadoop, today Yahoo released their own tested distribution of the code that powers their search engines, ad systems and webmail services. There is a growing move toward design patterns that leverage the parallelism inherent in distributed systems such as Hadoop and Google's MapReduce. Applications can be developed on single servers then deployed on massively distributed cloud infrastructure without knowing the details of such distribution. Once these applications are deployed they can act as their own service provider to other systems. Indeed we see that scenario with Amazon's EC2 (with native Hadoop support), IBM's Blue Cloud Initiative and Google today. This type of database is not a relational engine like Oracle or SQL Server. Hadoop enables large scale data mining for useful applications such as fraud detection and rich media indexing. I think this release is significant because it allows developers to take advantage of all the work put in to improve Hadoop over the years. Yahoo's change log file has over 8,400 lines and contains a wealth of knowledge gained by real production experience. Can Yahoo gain cloud credibility by giving it away? I think so; it gives everyone a living benchmark.