Before embarking on a large-scale server consolidation project, IT managers should carefully plan storage infrastructure and consider how best to optimize the infrastructure for both server and storage domains. As IT managers plan how best to incorporate server virtualization in their own environments, they need to address a series of storage-related challenges that, when taken together, can make or break a deployment.
- optimizing the storage layout to ensure a high level of application performance; and
- improving visibility, reporting and chargeback of storage in a virtualized server environment.
The first challenge involves ensuring that an organization's storage layout supports a high level of application performance. IT managers often focus on CPU utilization as the key determinant in how many application workloads can be consolidated onto a physical server.
But this is an oversimplification; memory utilization and I/O latency are also critical in determining application performance. Prior to consolidation, a typical server uses more than 60% of its memory, but just 10% to 15% of its CPU. Consolidating even two or three virtual workloads on a physical server can result in memory becoming overcommitted and overutilized. This also results in a high number of page misses and cache misses (both of which involve processor latency). These problems in turn lead to excessive page swapping -- another problem associated with processor latency -- and a highly randomized I/O pattern per logical unit number (LUN) volume, causing cache misses at the storage-array level.
To resolve this challenge, IT needs to optimize the storage layout to respond quickly to read I/O and page-level requests. Ideally, organizations should deploy storage systems with advanced storage virtualization capabilities that can process small-block, random I/O with low latency. To maximize both storage responsiveness and I/O throughput, companies should deploy systems that allow data to be striped and balanced across all available spindles and controllers. These approaches to storage virtualization will help ensure a storage layout that optimizes application performance.The next challenge for storage managers is how to improve the visibility of storage in a virtualized server environment. In Fibre Channel SANs, World Wide Names (WWNs) are assigned to physical servers, but not individual virtual machines, making it impossible for managers to gain end-to-end visibility from virtual machine to physical storage. Likewise, traditional storage tools do not monitor and report at a virtual machine level, making it difficult to trace allocation and utilization from virtual machines to volumes to physical disk.
To solve this visibility problem, storage managers should evaluate and implement virtual host bus adapters (HBAs). In a virtual HBA, a pool of WWNs is made available to each virtual machine, using a technique called N_Port ID virtualization (NPIV). NPIV enables the assignment of WWNs to individual virtual machines. This eliminates the need for multiple virtual machines to share a WWN and gives administrators visibility down to the virtual machine level. And with NPIV, as virtual machines move from physical server to physical server, the WWN moves with the virtual machine, which ensures continuity of reporting and storage management.
Once an organization achieves storage visibility at the virtual machine level, it can use SAN best practices and management tools, such as fabric zoning and LUN masking. Ultimately, IT managers will discover that the right mix of storage virtualization devices and approaches gives them the opportunity to achieve a high level of application performance and end-to-end visibility in consolidated environments. This end-to-end visibility also enables application-based policy management, including backup and chargeback policies. Virtual HBAs that incorporate NPIV functionality are available from such vendors as QLogic Corp. and Emulex Corp..About the authors: Steve Norall and Jeff Byrne are senior analysts and consultants at the Taneja Group in Hopkinton, Mass.