As organizations continue to scale their data centers to satisfy their business requirements, IT administrators invariably face the task of efficiently managing growth while ensuring maximum utilization of their data center resources. The current generation data center infrastructure, as shown in the image below, provides very limited options for efficient manageability and improving resource utilization across servers, storage and the SAN fabric. In addition, while the cost of processing power, storage and SAN ports are on the decline, operational costs such as cost of power, space, installation and administration are on the rise and are seen as a growing concern for organizations.
Virtualization-based technologies are addressing these concerns across all segments of the data center. Virtualization is the process of abstracting physical resources and presenting them in a logical form so that physical resources can be used in ways that are more efficient and manageable. This article explores the current limitations in the data center with a specific focus on SANs and discusses the virtualization trends in each of the three major segments of the data center - servers, SAN fabric and storage.
Limitations with Today's Server Deployments
In most data centers, a new server is needed for each new application that is deployed. While multi-tasking operating systems are capable of running multiple applications at the same time, limiting applications to one per server provides the advantage of fine tuning the operating system parameters for the specific application as well as preventing scenarios where a single application crash shuts down all other applications on the same machine.
The problem with the one-server-per-application approach is that servers are often underutilized. According to Microsoft and Sun Microsystems estimates, server utilization rates can range anywhere from 5 percent to 15 percent, which indicates that many computing cycles are left idle. If a company could efficiently consolidate five such servers each running at 10 percent or lower utilization onto a single server, then the company would only need to spend 1/5th (or 20 percent) of the overall cost. In addition, the resulting 50 percent server utilization after consolidation still leaves ample room for demand spurts.
Server Virtualization
One of the approaches that enable consolidation of servers is Server Virtualization technology. Server Virtualization allows the creation of multiple virtual machines on the same physical machine. Each virtual machine created then has its own set of virtual hardware (e.g., memory, CPU etc.) and is able to host an operating system of its own. This approach facilitates the use of multiple operating system instances and also makes it possible to run different operating systems such as Solaris, Linux, and Windows - all on the same physical server. By utilizing this isolated virtual server environment with its own operating system, multiple applications can now be run on a single physical server, thus improving the overall utilization of the server drastically. In addition, applications and their virtual servers can be dynamically moved from one physical server to another, based on load conditions - thus improving resource allocation and uptime of servers within the data center.
While server virtualization addresses the CPU utilization concerns through consolidation, the multiple virtual servers on the physical server are limited to a single view of the SAN regardless of their individual requirements. This limits storage management, QOS and resource accounting capabilities and gives rise to data protection concerns in the SAN. Vendors such as Emulex, are solving this problem by offering Virtual HBA technology that enables each virtual server to have an independent view of the SAN by utilizing virtualized HBAs and thus enabling customers to fully realize the benefits of server virtualization.
Limitations with Today's Storage Fabrics
In the data center parlance, the Fibre Channel storage area network (SAN) is referred to as a fabric. Organizations have a deployed base of multiple SAN islands or fabrics. These isolated fabrics are either a result of mergers and acquisitions or a result of internal SAN partitioning policies that were driven by application, technology migration or departmental isolation requirements. While a SAN by itself increases data availability, resource utilization and manageability, the existence of multiple SAN islands within an organization tends to decrease the overall efficiency of managing the SAN fabric. Separate fabrics mean more hardware, more ports, more devices to manage, and typically underutilized hardware. The ability to virtualize the various fabric resources such as the switches, ports and fabric services now enables organizations to gain the flexibility of assigning resources while still maintaining the required fabric isolation.
Virtual Fabrics
Virtual Fabrics is the technology developed to enable virtualization of fabric resources. This technology, standardized in 2004 through the INCITS T11 technical committee, enables isolation of SAN fabrics while also contributing to efficient fabric utilization.
The virtual fabrics approach enables the creation of multiple virtual switches and multiple Fibre Channel Services on top of the same redundant physical switch(es). Using virtual fabrics, storage network designers can improve the efficiency of a physical fabric and alleviate the need to build multiple physically isolated fabrics to meet their organizational or application needs. Spare ports within the physical fabric can be quickly and non-disruptively assigned to existing virtual fabrics thereby providing a clean method of growing application-specific SAN islands virtually.
While virtual fabrics in a "physical servers" only environment can be implemented without any support from the servers (this functionality is embedded in the switch port), in a virtual server environment, support is required from the HBAs in order to identify and group the respective virtual servers to their corresponding virtual fabrics. Vendors such as Emulex are addressing these requirements by providing the ability for a single HBA to simultaneously and natively access multiple fabrics. This enables virtual servers to be uniquely identifiable in a virtual fabric, thus allowing virtual servers in a single physical server to be grouped into different virtual fabrics based on application or workload requirements.
Limitations with Today's Storage Deployments
Similar to inefficiencies in physical server utilization, storage arrays also face concerns related to inefficient resource utilization. In order to maximize the utilization of various disk drives within an array, vendors have implemented storage virtualization at the array level. While the individual storage arrays are efficient at managing the storage within the array, capacity utilization falls significantly when considering all the arrays in the entire fabric. A single array that is nearing its maximum capacity utilization could force the need for a new storage array even though there are other arrays with very low capacity utilization. If a company could consolidate their various storage arrays into a larger pool of shared storage, it would increase storage utilization, and thus avoid the costs associated with purchasing additional arrays and related ports.
Network-Based Storage Virtualization
Network-based storage virtualization enables pooling of physical storage from heterogeneous storage arrays into what appears to be a single storage device that is managed from a central console. It also permits storage resources to be altered and updated on the fly without disrupting application performance, thus drastically reducing downtime. Other key advantages of network-based storage virtualization include:
- Easier migration of data between physical storage resources - enabling tiered storage architectures for information lifecycle management
- Support for intelligent applications that solve requirements such as data availability, disaster recovery and point-in-time images for testing and data mining
While in-band network based virtualization appliances have been available for three to four years, the emergence of intelligent storage platforms has enabled network based storage virtualization deployments with split-path architectures. These intelligent platforms that have taken shape in the form of a switch or a dedicated appliance provide the necessary hardware acceleration for storage virtualization functions and significantly improve the scalability and performance of network-based virtualization solutions. Vendors such as Emulex are addressing the needs of these intelligent platforms by providing intelligent storage processors that are built into various platforms including storage appliances, intelligent switches or front-end arrays. Further, the inherent performance and scalability benefits associated with intelligent storage processors typically tend to eliminate the debate around which architecture to deploy.
Improved Data Center Efficiency Through End-to-End Virtualization
The technological developments in server virtualization, fabric virtualization and storage virtualization are set to usher in a new paradigm in the enterprise data center. With the capability to create virtual servers, virtual fabrics and virtual storage on demand, the time taken for provisioning the infrastructure required for data center applications can now be significantly reduced. More importantly, this will enable IT administrators to create an end-to-end resource profile for new applications (CPU, memory, fabric bandwidth, array quality, backup frequency, etc.) and let automated processes instantiate, maintain, and reallocate the required resources even as physical resources are dynamically added, removed or reassigned.
In addition to dramatically increasing flexibility in managing the data center resources, end-to-end data center virtualization will also improve the operational efficiency of the data center by lowering the cost of operations through consolidation of resources and by offering increased levels of service assurance enabled by support for high availability and automation.
In all, an end-to-end virtualized data center will provide a utility-class infrastructure for offering a new range of services. Combined with computer and storage grids that enable better deployment at the physical level, data center virtualization will provide for a service-oriented architecture or SOA.
Over the next few years, virtualization will transform the way IT approaches the data center. Virtualized systems, such as Emulex end-to-end data center virtualization products, will be the default configuration of the next generation data center.
Ranga Bakthavathsalan is Senior Product Marketing Manager, Emulex Corp.
No comments:
Post a Comment