Server Virtualization: Memory Serves As Cost-Effective Option

By Marshall Shah
published: Monday, August 04 2008

Companies of all sizes around the world today are actively working on server virtualization projects to reap the benefits of consolidating underutilized servers. By reducing the number of servers delivering the same workload, IT organizations are reducing their space, power and cooling requirements, as well as the total cost of ownership (TCO) of their servers, while increasing their server return on investment (ROI). As a result, IT administrators are faced with having to transition workloads from a purely physical environment to a new virtual environment.

Business Goals and Challenges

As IT organizations have gained experience with consolidating many physical servers onto one server, many are finding that consolidating onto older servers with fewer/lower-power processors and support for smaller memory sizes does not meet the performance needed for the workloads and limits the desired consolidation ratios of the data center. This disparity between desired and actual results is leading to the purchase of new servers and/or blades with increased CPU performance, increased memory, increased I/O connectivity, and the use of networked storage, usually a SAN. In practice, the limiting factors for the configuration of memory include:

• The processor architecture (32-bit vs. 64-bit)

• The number of memory slots available

• The cost of high-capacity memory

Goals

• Achieve high consolidation levels

• Increase efficiency and flexibility of computing environment

• Maintain optimal application performance to meet changing business requirements

• Reduce costs in areas including hardware, support, and power

Challenges

• Able to run many workloads in VMs with good performance

• Specify memory requirements to accommodate the need

• Poor performance due to oversubscribed memory 

• Accurately determining the optimum memory size

Virtualization and Memory

It is worthwhile to take the time to analyze and understand how virtualization solutions use memory, how much memory overhead each virtual server and associated workloads require, and the ramifications of both over-committing and over-purchasing memory. There are four areas where memory may be required in a virtual server implementation (see Figure 1):

• the memory required to load the hypervisor (typically a small amount)

• the memory required to initiate a virtual machine

• the memory required by the operating system loaded into the virtual machine

• the memory needed for the application

Figure 1: Components used in a virtual sever environment that require memory

It is also important to consider the various available options for populating these consolidated systems with memory and building a financial justification for purchasing the appropriate amount of memory.

Add Host or RAM

One of the critical questions that many IT personnel wrestle with is whether to add a host or add more memory.  If your server CPU has capacity, adding RAM whenever possible may be the best option.  Clearly, there are benefits to adding more RAM over adding a new host.

Adding a Host requires:

• Licensing, rack space, power, and cabling

Adding RAM requires:

• No licensing, no space, minimal power, and no cabling

Adding a virtual machine requires management and access to storage (which may or may not be already available). Due to the fact that the virtual machine is sharing the power, cooling, space, cables, and maintenance with other virtual machines on the same server, the total cost of ownership for a virtual machine is significantly lower than it is for a physical system. The ability to add more high-quality memory to a system in order to support more virtual machines and the associated workloads avoids much of the costs associated with provisioning a new "server," thereby lowering the total cost of ownership for supporting those same workloads in the data center.

Return On Investment

The basis for the return on investment for adding memory to a physical system will be realized immediately in cost avoidance. The ability to simply add memory with a list price of less than $100/GB to support and grow the number of virtual machines supported avoids incurring the expense for another system with associated maintenance and support fees.

The systems from HP and IBM that are used for this example come with 8 GB of memory standard. To support the 10 virtual machines with 10.3 GB of memory, either additional memory or another entire system must be purchased. Table 1 outlines the ROI for adding 64 GB of memory to the machines rather than adding an additional system.

Table 1: ROI for adding 64 GB of Dataram memory

*Prices based on June 2008 figures; prices subject to change

Recommendations

Virtualization solutions allow IT to consolidate workloads from many underutilized server resources onto one system. To achieve both the desired consolidation ratios and adequate performance, following are some recommendations to consider:

• Where possible, maximize the consolidation ratios and the return on server investments by adding more memory, rather than purchasing another system.

• Minimize the total cost of ownership through avoiding the power, space, cooling, cabling, and maintenance associated with another system.

• When purchasing additional memory, consider using third-party vendors for high-quality, cost-effective options.

Dataram , HP , IBM