Home > Blogs > Rethink IT > Author Archives: John Gilmartin

Author Archives: John Gilmartin

EMC ViPR: A New Storage Platform for the Software-Defined Data Center

Today at EMC World, EMC announced ViPR, a new open storage platform that enables the abstraction of the storage layer into a single pool of virtual storage within a software-defined data center. ViPR will easily integrate with VMware-based environments, and will enable organizations to centrally access and manage EMC and heterogeneous physical storage infrastructure.

By extending the benefits of the software-defined datacenter to storage, customers will be able to use their existing VMware infrastructure with ViPR to drive greater value, automation and simplicity out of their existing storage solutions.

At VMware, our mission is to extend the benefits of virtualization to all areas of the data center — beyond compute to security, networking, management and storage. With the help of our strategic partners, such as EMC, we can help customers realize greater efficiency, flexibility and agility in their IT infrastructure through a software-defined data center architecture.

Please see the EMC news release for further details on this announcement, and also a blog post on the news from Amitabh Srivastava, President, EMC Advanced Software Division.

Can Applications Manage Themselves?

John Gilmartin here, Director of Product Marketing for Private Cloud products and solutions.  I just left the Gartner Infrastructure Operations and Management conference where I attended a fascinating session on Cameron Haight’s research into public cloud providers.  He contrasted their architecture and management methods with those of the typical enterprise IT organization.  He covered a lot of ground but some key ideas that stuck with me:

  • Cloud providers achieve server:admin ratios of 1000+ while most enterprises are less than 100:1. 
  • They achieve this through a focus on removing complexity and by empowering small teams.  They foster close interaction between operators and developers with “infrastructure as code” as a critical concept – applications are built with the infrastructure in mind. Public cloud operators experiment often and focus on improving those processes with the highest impact.  This all reminds me of the Toyota Production System, a highly efficient, team-oriented improvement methodology, focused on continuous learning and improvement. 
  • For Enterprise IT, systems management is usually an afterthought with management systems bolted on after applications are already deployed, adding complexity on top of complexity.  Meanwhile, managers enforce top down processes around “best practices” – think ITIL.  Cameron pointed out that these methods are rooted in Frederick Taylor’s early 20th century approach to management so popular with American car manufacturers until the 80’s.  (Remember what happened to American car manufacturing in the 80’s?).

As a practical example, it occurred to me that the complexity of implementing traditional Application Performance Management (APM) systems illustrates this well.  To build an APM solution, you map out the components of an application, define the dependencies, and collect a whole series of indirect measurements like processor utilization.  You populate all of this into a service model with a big historical database, spend a long time tuning the thresholds, and then dig through a bunch of charts to try and solve problems. 

Great in concept, but this service model replicates the complexity of the underlying application, yet is not intrinsically tied to the application.  When the business service changes (which happens often) the model must be updated.  Either you experience “model drift” as people forget to update the model in their haste to support the business, or you put in place bureaucratic change processes that slow the speed of business. This approach falls short in the traditional IT environment, so how can it possibly work in the exponentially more dynamic cloud environment?

Is there a better way?  In manufacturing, you have the concept of “design for manufacturability (DFM)."  Do Enterprise developers need to “develop for management” by explicitly designing for operational management when building the application?  Can modern application frameworks like Spring make this easy for developers?  Can we at least assign outcome-based policies to applications and allow an intelligent infrastructure to automatically manage to those policies?

Share your opinions and let’s start a discussion on the future of management.