The Datacenter of the Future – What Was Once Old Is Now New Again

Design Editorial

Thursday, 13 November 2008 14:28

Written by Jon Greaves

Timesharing eventually gave way to client-server computing as users looked for richer graphical interfaces and realized the economics of moving to much smaller servers costing significantly less than mainframes. The 90’s and the new millennium ushered in distributed, web-based applications that further distanced companies from a centralized, Service Bureau computing model, introducing  the reemergence of an old trend … the rise of centralized, shared computing resources.

Going back to the 60’s and 70’s, Service Bureaus published access to their applications over dialup or packet switched networks and used time-sharing capabilities of mini or mainframe computers to allow multiple customers simultaneous access. This proved to be a successful business model when customers evaluated the expense of purchasing their own mini-computer, installing the applications, then running jobs vs. renting capacity on an existing system. As a result of users renting applications by time-slices, the utilization of mini/mainframe computers increased, making this a very natural solution for organizations with available computing power or dedicated Service Bureaus.

With the advent of advances in networking technology, desktop computers became more powerful and readily available allowing this time-sharing approach to evolve into client server computing. Applications can now be distributed between a desk-top computer and a back-end server. 

Client server gave way to three or n-tiered where the business functions were extracted from the applications user interface or data. This allowed the functionality of applications to be easily extended without necessarily having to recode the user interface or database back-ends.

Fast forward to today’s datacenter. With the advances in virtualization technology now providing the equivalent of time-sharing not only for computing resources, but for storage and networking, we see the re-emergence of the Service Bureau model as Software as a Service (SaaS) vendors. These vendors publish their applications in much the same way as the original Service Bureaus.

While SaaS vendors focus on their application being delivered in this model, a number of companies have generalized this capability, allowing you to run your own applications in this mode with little or no modification. This technology has been labeled with many names ranging from platform or infrastructure as a service, to the overused (and abused) cloud computing. 

These solutions offer great advantages to customers with workloads that fit this model, typically including lower service level agreements (around 99.9% availability) and those comfortable with customer data being hosted in such a shared cloud. Early adopters of such technology are the “Web2.0” startups that do not see strategic value - at least at the current point in their evolution - to run/operate their own infrastructure or those using such technology as a tier in their computing or storage strategy.

The benefits of adopting a highly virtualized computing platform have also caught the eye of many enterprises and government agencies. While they may not be able to adopt a shared cloud solution, they do desire the potential savings such a solution can provide. This area is evolving to what’s basically a private or semi-private cloud, with the latter being infrastructure shared between multiple divisions or partners, providing compatible availability and security profiles.

Some customers are creating tiers of computing. By blending both models of shared and private clouds, and selecting a private infrastructure for core applications requiring a greater degree of control of availability and performance, they can then leverage on-demand shared cloud services to provide burstable capacity. 

A great example of this is SmugMug - a photo and video editing, sharing and printing site.  SmugMug uses Amazon’s S3 service to store in excess of 500TB of digital photographs and videos. For its compute platform, SmugMug uses a combination of its own hosted servers and dynamically provisioned computing from Amazon’s EC2 compute cloud. SmugMug has taken this model to the next level by designing an autonomous control system, called SkyNet. In essence, SkyNet watches for specific workloads that can be moved to Amazon’s compute platform such as the encoding of video or the application of watermarks to images. This can equate to the provisioning of several thousand virtual servers to complete their workloads. As the demand is met, these virtual servers are then decommissioned - all automatically driven based on demands and service level agreements.

This is a far more cost-effective method of meeting customer demand, where the alternative would be purchasing and provisioning thousands of servers that may only become utilized once a week.

As customers begin to embrace these forms of dynamic computing, the demands on managed hosting providers as well as what will be required to support these forms of computing will evolve. A new breed of datacenter optimization services focused at the customer’s workload will augment today’s smart hands and basic managed services. 
  

  

A good example of this is capacity planning. Today many enterprise customers consider capacity planning a quarterly, or potentially annual event linked to purchasing cycles of new hardware and square foot/power from hosting providers. 

In the environments previously described, the discipline of capacity planning evolves to become “capacity prediction” – a constant activity where the customer works closely with the managed service provider to ensure the most appropriate computing power is available when they need it. “Power” is the relevant term here. With the cost of electricity spiraling to be 60-80% of an organization’s IT costs, customers are looking for opportunities to optimize their power usage. In fact, power is such an important factor, many customers are now investigating or deploying “follow the moon” solutions where applications are migrated to follow the inexpensive electricity typically available from energy companies from midnight to 5:00am. Another door opened by highly virtualized environments and requiring close partnerships with managed service providers.

So what does all of this mean to hosting providers? The last data center construction boom occurred during the dot-com bubble nearly ten years ago. Many of today’s datacenters were constructed with a shelf life of 10-12 years with the expectation of retrofits required to catch-up with power and cooling demands of modern infrastructure.  

Let’s consider the kinds of servers these datacenters were designed for. These “enterprise” class servers typically required 610 watts of power for a four-rack unit machine. In most cases rack space would be exhausted before the power. If you contrast this with today’s modular blade servers, we see two to three times the density as compared to traditional rack mount systems where a fully populated chassis running at high load can require over 6500 watts in just 19 rack units. 

From a data center point of view this changes the power provisioning equation from being 100watts per sq. ft. to areas requiring greater than 300watts per sq. ft. putting strain on datacenter power plans and utility companies to provide the electricity. This often leads to datacenters running out of power and being part vacant.

As power consumption increases so does the heat required to be dissipated leading to new blueprints for construction. We are seeing designs move away from raised floor paired with cooling the datacenter itself using computer room air conditioning (CRAC) units to hot aisle containment and in rack cooling based on concrete slab construction.

As hosting providers begin this refresh cycle they need to consider both the changing workloads and new services customers will undoubtedly require. The datacenter of the future will tie the changing workloads along with intelligent facilities to allow the customer to manage risk, price and performance of their workloads. These workloads will become highly portable by design and capable of spanning datacenters - and potentially continents - to deliver the value of dynamic computing customers require.

Interested in learning more about the datacenter of the future? Join us for a LIVE webcast Wednesday, December 3, 2008 at 11:00am EST.  To register for this event click here. Once we receive your registration, you will have the opportunity to download the first installment of our e-Book, “Datacenter of the Future” as well as a brief summary of the first three chapters.

About the Author: Jon Greaves is a recognized leader in the information technology services industry, having spent a significant part of his career in managed services and operations with a particular emphasis on remote services delivery. Most recently as CTO and Distinguished Engineer of Sun Microsystems Services business - and prior to that - Jon provided an instrumental role in the success of SevenSpace, a pioneer in remote, IT operations services.  Jon is an expert on systems security and privacy, playing a strategic role in representing the telecommunications industry in developing international standards in response to Critical Infrastructure Protection and U.S. Presidential Decision Directive 62 and 63.  Jon has also held positions at British Telecom, MCI and Concert.For more information, visit the Carpathia Hosting website at www.carpathiahost.com. 43480 Yukon Drive, Ste. 200 Ashburn, VA  20147 - 703.740.1730 (voice)703.997.557 (fax)

TrackBack URI for this entry

Comments (0)

Subscribe to this comment's feed