The Lippis Report

Information demand is growing faster than Moore’s Law. IDC recently predicted that between 2008 and 2012 IT staff will grow at approximately 1.1 times the rate of business growth while servers will grow at 1.9 times, mobile internet users will grow at 3 times, non-traditional user devices will grow at 3.6 times, information will grow at 4.5 times and interactions per day will grow at 8.4 times. Clearly, the gap between IT staff resources and business expectations for IT services is huge and growing fast. To close the “business expectation” gap many IT leaders are evaluating public cloud-based services to augment their private data center/cloud services as IT leaders observe an increasing amount of applications being deployed from outside the enterprise perimeter. Salesforce.com, EC2, Google Apps and other cloud services will have a huge impact on their data center capacity. The enterprise network is an integration point in an IT architecture allowing homegrown, private and public cloud services to be deployed securely and reliably.

IT business leaders have long been seeking more dynamic data center infrastructure that allows applications and services to scale with demand, both up and down. This flexibility of service delivery has become acute during the recent economic downturn as business leaders cut under performing business models, streamlined business processes and sought to quickly enter markets as they developed. This business agility requirement to usher in new business priorities and processes need prompt data center provisioning and service delivery automation. The time to provision a server, storage, network and application is too long, often measured in months. Provisioning automation is paramount in the data center as it quickens the pace of business while also contributing to business continuity initiatives.

In addition to scale, many IT business leaders are confronted with the challenge of greening or reducing their data center power consumption and cooling requirements as they represent approximately 25% and 15%, respectively, of total amortized data center cost. For each watt delivered to a data center approximately 59% is consumed in IT equipment, 8% to power distribution loss and 33% to cooling. Over a 3-year period energy cost can be twice server acquisition cost! In addition, data centers or more accurately server utilization can be very low, as low as 10%. Making matters even worse is that servers draw as much as 65% of peak power while idle. Two key design goals are to increase server utilization to near 35% and reduce energy consumption so that the Power Usage Efficient or PUE (Total Facility Power/IT Equipment Power) is 1.7. Note inefficient data centers run at 2.0 to 3.0 PUE while extreme Green data center projects are striving for an ultra low PUE in the 1.05 range.

The above business pressures placed upon data center resources are in addition to operational pressures that most are confronted with. Increasing productivity is a top priority. Well-run data centers typically have a staff to server ratio of 1:1000. As productivity pressures increase so too do service levels. Service Level Agreements (SLAs) between IT and business units are changing data center business models, as IT becomes more of an internal service provider with formal chargeback mechanisms for use of infrastructure services. Therefore, IT business leaders are challenged with increased efficiency, productivity and service delivery.

Virtualization Trends

Data center virtualization has emerged as a main solution to address many of the challenges identified above and has become a means to enable large-scale data center consolidation. As CPU suppliers move to multi-core processors and server virtualization offers multiple isolated application and OS pairs per server, IT leaders are able to build larger scale data centers measured in the 10s to 100s of thousands of servers. As more applications are loaded upon servers their utilization increases too, reducing power consumption, cooling and server acquisition requirements. Because of virtualization’s value it has become a “top down” executive management decree as companies that embraced it early realized business value in controlling and reducing operating costs, easing sub-function spent and extending the life cycle of acquired capital assets.

Why Massive Virtualization is Inevitable.

IT service delivery demands are growing faster than Moore’s Law which predicts a doubling of transistor density every two years and in today’s engineering that’s a doubling of CPU core width. This leaves IT leaders to either re-write applications for multi-core processors or settle for halved deficiency every two years. The path of least resistance has become to virtualize servers, but this too offers challenges around span of control. Virtualization has fundamentally changed the role of IT executives in charge of networking, storage, servers and applications. IT organized around competency centers, or centers of excellence are finding that responsibilities around IT silos are changing permanently.

In short, server configuration changes now impact networking and storage. Virtualization breaks the premise that IT can be organized around isolated technology groups. Server virtualization is powerful enough to IT business leaders that they are willing to sacrifice the 15-year proven network design approach to achieve business benefits. Therefore, networks are changing, servers are changing and storage is changing to embrace and accelerate virtualization.

Hybrid Virtualized And Non-Virtualized Data Centers

But for all the value associated with virtualization there are impediments to deployment. For example, not all applications are capable of running on a VM. Many IT leaders and application teams struggle with Microsoft Exchange, SAP and Oracle database implementations, for example in virtualized environments. In the mean time, some IT groups are deploying Oracle, for example, on non-production virtualized machines to gain experience and confidence. But many older or legacy applications cannot run on a virtualized machine or they have not been tested in a virtualized environment on various blade offerings. IT leaders are concerned with professional service cost to port legacy applications onto a VM while application teams are slow to fully embrace virtualization due to performance and reliability fears. The end result is that most servers are not virtualized and the industry is slightly beyond the early adaptor cycle.

The de-facto standard is that the majority of data centers live in a hybrid state of virtualized and non-virtualized servers. While this hybrid state will last for many years to come the trend line is clear that the portion of virtualized servers will increase, and increase significantly, especially for x86 servers, over time. In fact x86 servers are typically on a two to three year refresh cycle, which IT business leaders are using to synchronize their vitalization plans. In short, new x86 services will be increasingly virtualized.
The virtualized side of the hybrid data center will grow, as application teams feel comfortable about virtualization. While there are many server virtualization providers such as VMware’s vSphere, Microsoft’s Hyper-V, Citrix’s Xen virtual servers etc., VMware dominates the market segment. Its vSphere 4.0 goes a long away to increase performance, management, security and visibility capabilities of virtualized environments, a major concern of application teams.

vSphere adds significantly more power and flexibility as current ESX hypervisors only support up to 64 GB of main memory allocated to a single VM on a server and can only span up to four x86-64 cores. vSphere will span up to eight cores and address up to 256 GB of memory. In addition, vSphere allows application teams to change the amount of RAM allocated to VMs without rebooting. These capabilities are going a long way toward making application teams conformable, as it should help avoid disruption or downtime when making a memory change. The new vSphere’s maximum RAM limit is now set at 1 TB. Furthermore, an integrated Microsoft PowerShell command-line interface can be used to adjust the configuration of a VM running Microsoft Exchange on the fly. Many of these features are solutions to key technical hurdles, such that application teams will now be able to virtualize larger, more-mission-critical applications, such as Oracle, SAP and large Exchange implementations.

With server virtualization an efficiency technology compacting operating systems and applications into every increasingly more powerful server hardware, IT leaders are afforded lower cost, greener data centers and most importantly scale. Cloud spec data centers are able to scale to 100s of thousands of servers thanks to virtualization. With all this scale and increased network speed in the 10 Gb to 40 Gb and soon 100Gb a new approach to storage and networking is afforded that combines their access over a single fabric. This converged I/O via a single NIC card that splits storage and network traffic further reduces data center energy consumption, cabling and equipment cost while contributing to workload mobility requirements by automating network and security changes required during such moves. But these advances are not just afforded to public cloud providers; they are available to IT leaders building private cloud too. In fact, IT leaders are moving just as fast as cloud providers to take advantage of these new data center design approaches especially as large IT suppliers such as HP, IBM, Cisco, Dell and others offer new blade system designs that package computing, virtualization and converged networking into units.

In upcoming Lippis Reports we’ll focus on these new blade systems and converged network designs that offer new economics and performance advances.