VMware: Under the Influence

By Bob Maness
published: Tuesday, October 28 2008

Data center resources have traditionally been underutilized while drawing enormous amounts of power and taking up valuable floorspace. Virtualization has been a positive evolutionary step in the data center, driving consolidation of these resources to maximize utilization and power savings, as well as to simplify management and maintenance. However, adding virtualization to the picture forces servers to run multiple applications with partitioned resources, pumping more I/O to the storage systems. This causes much more strain on a storage system than the same amount of non-virtualized servers.

There are many entrants into the server virtualization market; however VMware is the leader in server virtualization, enabling data center operators to do more work with fewer physical servers, while driving up server utilization numbers to unprecedented levels. It enables administrators to run more workloads on a single server, and facilitates virtual machine mobility without downtime. With all of the benefits of VMware comes a wholesale change in the way you must plan for backup, restore, and recovery.

Virtualizing Servers

To fully benefit from virtualized servers and the ability to instantiate virtual machines on demand and move them from physical server to physical server, it is essential to have a consolidated and virtualized set of storage resources available to the VMware environment. There are several reasons why it is beneficial to connect virtual servers to consolidated networked storage and not rely on storage directly attached to each physical server. If a group of servers is virtualized, as is often the case, it is more efficient and economical to connect them to a shared storage resource - and this means networked storage.

Leading industry analysts have found that server virtualization has required a net increase in total storage capacity. Companies often end up purchasing double the amount of storage required when servers are virtualized. They do this in order to alleviate the decrease in utilization and performance associated with virtualized server projects. For example, a physical server running five virtual machines imposes up to five times the burden on its storage that it did before it was virtualized. The virtual machines suffer from back-end storage contention because there is a server-to-storage I/O bottleneck, particularly evident in high-end server consolidation projects. For many users, this is an unexpected and unpleasant surprise as well as a hindrance to the adoption of virtualized servers.

Application-Aware

A LUN from a SAN-attached array is treated equally by the array. Most virtualization software will simply place their files on the logical unit number (LUN) or make some kind of raw access meta-data file, so this makes tuning the array for virtualization pointless. According to the best-practices guides from many virtualization vendors, the value of write cache, read-ahead, and all the value add of the high-end array go out the window.

But what if there was a way to assign array resources (CPU, cache, and disk) based on the applications that will be using them? The answer is application-aware. This would improve the array's value by greatly increasing its utilization as well as increasing the performance of the applications that run on it. The result would be better server consolidation, resulting in cost savings. By using a an application-aware storage system, physical servers can be more efficiently virtualized.

Additionally, an application-aware system can dynamically assign storage to improve performance for the applications deemed most important, allowing an increased number of virtual machines to be placed on a physical server by clearing up back-end storage contention. The result is an array that shows the same resource preferences as your servers. This improves performance for the applications to which you assigned greater server resources, while not punishing the workload from other applications. The application-aware system will treat I/O from your provisioned application with a greater priority, giving it more cache and higher performance than other applications that you have deemed to be less critical to the business. This helps reduce or eliminate the server I/O bottleneck in high consolidation projects and makes better use of today's quad-core servers. By managing storage resources based on application needs, an application-aware system can provide performance and utilization gains compared to traditional storage systems.