Data Center Automation and the Agile Enterprise
The emergence of application-ready, modular utility computing platforms
by Mark Sigal
Envisioning the Agile Enterprise
In the Agile Enterprise model, application-ready software stacks can be:
- moved seamlessly between physical and virtual computing resources.
- provisioned on one or more systems dynamically.
- autonomously updated and patched according to user-definable compliance and security policies.
- scheduled, executed and tracked according to logical sequences, events, dependencies and geographic hierarchies.
To set the context for this new model, consider the process of provisioning an e-commerce application. The steps in such an exercise might include first launching the database; next, verifying that the database is running optimally and defining polices for dealing with suboptimal conditions. Once that checks out, then the application server could be launched and a verification path specific to the application server could be completed. After these steps are successfully completed, the application server must be pointed to the database; the Web server must be launched, and so on. And this is a relatively simple application. Replicate this model across dozens of applications and hundreds, or even thousands, of servers—running across geographically dispersed locations—and it becomes even more complex to manage. Factor in clustering software and load balancers to accommodate scale and continuity, and the situation becomes positively unwieldy and human capital-intensive.
What makes managing such an application so complex? Consider that all the steps articulated above have enormous dependencies that require appropriate actions and responses to events. One such event is the response to a system failure or the eclipse of a performance threshold that compromises service-level objectives. Some of these actions, such as ensuring proper patch installation, involve both dependencies and calendar-based scheduling criteria. This can be complicated. One application server may run best with a certain patch, but the operating system must have a different patch installed for events to work at all in this configuration. From a scheduling perspective, updating 500 systems in the middle of a workday is out of the question, so start and stop times need to be defined for a given task.
Moving from Manual and Passive to Active Automation
Today, the absence of mechanisms to automate these tasks means that teams of IT personnel must work on weekends to roll out a new update. This occurs, in part, because of the lack of visibility into the progression of tasks or support for rollback methods. An established rollback method would restore the system to a prior working state when tasks fail (or are only partially completed). This entire state of affairs relegates data center automation to alchemy or blind trial and error. In fact, the “state-of-the-art” for automation in most IT shops is script libraries written in PERL, PHP and/or Python. Few data centers can even dynamically track what hardware and software assets are installed. Things have degraded to the point that companies often lose count of how many software licenses are “deployed” relative to “available” licenses. As a result, companies often report spending tens of thousands of dollars on unnecessary extra copies of licenses.
Most managers would be shocked to know that such critical asset information lives in spreadsheets or comparable static repositories. Can it really be this bad at billion dollar enterprises? The sad answer is “yes.” In fact, a couple months back one of the VPs of infrastructure at a global banking powerhouse lamented that their CEO does not understand why they can only “approximate” how many servers are installed at any given time. Is it any wonder that the typical IT manager can only administer 25 to 35 systems? Is it any surprise that 70 cents of every dollar is spent on maintenance versus innovation?
What is required for an active automation model to work in the real world? First, IT managers must embrace modular computing platforms. What are they? Modular computing platforms are systems built around Intel-architecture CPUs, Windows and/or Linux that leverage standard commodity networking and storage subcomponents.
Adopters of such platforms generally opt to remain platform-agnostic whenever possible. Even if they favor a particular hardware vendor, the manager strives to maintain flexible sourcing options. This eases transitions among operating systems, physical and virtual configurations, as well as standalone and clustered deployments.
Compelling Economies of Scale
Put another way, modular computing platforms are the successor to proprietary hardware clusters. While less powerful and less richly integrated than proprietary hardware systems, modular platforms are available at a fraction of the cost. The basic premise of modularity is that you can throw many $2,000 servers at what used to be a $1M high-performance computing problem.
Such an embrace is the proverbial tenfold increase in price-to-performance, as one hundred of these servers offer significantly greater performance than a single system at a fraction of the cost. Moreover, because modular systems are effectively standardized, there are a wealth of open standards-based utilities for provisioning operating systems and applications, installing patches, and doing recurring tasks such as configuration management. Standards such as Wake-On-LAN, IPMI, PXE, WMI, LMI, SSH, ADS, RPM, SNMP and CIM provide the essential building blocks for bootstrapping systems from bare metal. They can install necessary software, initiate configurations, and manage system, network and storage resources.
The emergence of standards in these critical, recurring areas of systems management is vital to widespread adoption of the Agile Enterprise model. The reason standards adoption is so vital is that a potential downside for modular computing platforms exists. Adopters of this model generally find that the number of server instances under management increases by several hundred percent. This happens because high performance systems are being replaced with cheaper ones. Given shrinking IT headcounts and current server-to-administrator ratios, marrying the standards to next-generation data center automation architectures is the key to the model’s viability. Without automation, the leap to modular computing platforms is not possible
Modular Computing Platform with Application-Ready Software Stack
Application-ready software stacks are built on top of the modular approach to hardware and system manageability. The software stacks also leverage a common set of underlying standards to bring comparable manageability options to the software layers of a system. In the case of the open-source, application-ready software stack known as LAMP—Linux, Apache, MySQL and PERL—IT operators have as much granularity of control as desired, since all of the application programming interfaces are exposed in open-source applications.
Furthermore, since these applications can run unmodified on almost any modular hardware platform, moving applications from less powerful systems to more powerful ones, adding them to or removing them from clusters, or transitioning them between physical and virtual server resources is largely friction-free.
Workflow as Workhorse
The true workhorse of an Agile Enterprise, however, is the workflow-based automation system. If modular computing hardware platforms provide the fertile ground for growing an Agile Enterprise, and application-ready software stacks are the seeds that can be planted according to real-time business needs, then the workflow system is the tractor that tills the soil tirelessly, powerfully and efficiently.
Workflow-based Automation Model
By this I mean that, when one peers into a typical data center environment, a number of recurring management processes are occurring at any given moment. New systems are being added to the data center. Trouble tickets are being opened, and efforts are undertaken to determine the root cause of such problems. Security policies are being monitored and enforced, and vulnerabilities are being assessed. Moreover, in many industries, regulatory compliance must also be tracked. Further, virtualization and on-demand computing models introduce a whole new set of processes that must be orchestrated and synchronized, often in parallel. Finally, disaster recovery sequences, while needing to be called on infrequently, are also highly procedural and thus prone to error in absence of a robust automation system.
While the above addresses the core of what is involved in maximizing infrastructure availability and uptime, the goal of IT personnel is to be better aligned with the evolving needs of the business organization. This implies a flexible, adaptive model for deploying new services on top of the infrastructure. Here again, the workflow-based system can be called on to automate deployment, activation and optimization of simple applications to complex, distributed services. Furthermore, as new voice-based data and VOIP services running on standard computing platforms take hold in the enterprise data center, operators have to manage higher levels of reliability and quality of service (QOS) than ever before. If architected properly, a workflow-based automation system can enforce QOS policies across hardware, software and networking resources.
ITIL Compliance and Best Practices
A real benefit of workflow-based automation is its synergy with ITIL (IT Infrastructure Library), the process-based methodology of IT service management. After all, a workflow-based system can support sophisticated process modeling and process execution, since it explicitly anticipates schedules, sequences, events, dependencies, and even distance! Further, since it incorporates a dynamic repository of assets (represented via an XML/CIM-based, unified management data access model), the workflow-based automation model can be readily extended to support standards for business process execution, such as BPEL4WS (Business Process Execution Language for Web Services).
One caveat should be noted regarding the incorporation of workflow-based technologies into a data center automation strategy: Many workflow engines are out in the commercial market and the open-source realm. This prompts the question of whether any workflow engine is a decent building block for data center automation.
The short answer is “no.” It is important not to confuse the automation ingredient with the automation recipe. In the context of data center automation, a workflow engine must be able to proactively monitor and asynchronously receive system faults and performance threshold events. It must also be able to speak the “protocol soup” of the myriad standards for system wakeup, bare metal OS and application provisioning, patch management and configuration management. The workflow engine must have a built-in mechanism for dynamically tracking what is installed and where. And it also must have the intelligence to be able to communicate through firewalls, across subnets and geographic locations. Finally, it must be able to mediate between disparate protocols, management interfaces, and access methodologies, and communicate with legacy scripts. Whether you choose to buy such capabilities or build your own, the key is to begin with the end in mind!
Turning IT from a Liability to an Asset
The promise of the Agile Enterprise computing model is irrefutable. You can realize a ten-fold price to performance benefit by treating computing platforms as commodity components. Free your IT people from working on the tedium of low-level tasks to working on perfecting high-level business processes. In doing so, you can create an application deployment model that is highly adaptive, increasingly automated, and one that shifts more of the IT budget from maintenance to actual innovation. The technologies to accomplish this are mature and the standards are fairly ubiquitous. Making the leap from passive and manual to active and automated is more of a mind change than anything else. Remember, each wave of computing not only leapfrogs its predecessor, but also fundamentally changes the rules of the game, creating new opportunities for growth and prosperity. There is no better time to get started on building an Agile Enterprise.
Mark Sigal, chief executive officer, UXComm
Mark Sigal brings 15 years of entrepreneurial experience to UXComm. Mark has been affiliated with six startup companies, including Rapid Logic, a unified device management company. As founding CEO, Mark led Rapid Logic to the position of fifth fastest growing company in the Bay Area.
Prior to UXComm, Sigal was a cofounder of Verdada, an Internet dashboard software startup. Sigal previously ran The Middleband Group, a strategic product-marketing consultancy that provided global strategy, product marketing, product management and marketing communications services to early and mid-stage startups.
A graduate of UCLA, Sigal is a frequent public speaker and writer. He also maintains a Web log for O’Reilly & Associates.