An expensive part of the virtual infrastructure is the physical servers that act as hosts to virtual servers or desktops. These physical systems, called hosts, are often ordered with maximum amounts of CPU horsepower, system memory and I/O ports to support the virtual machines (VM) that will be their guests. Since these physical servers which act as hosts are significantly more expensive than the single purpose servers they replace, a key to achieving a high return on investment (ROI) is to make sure these physical systems are hosting the maximum number of VMs, meaning that no part of their resources is underutilized.

The challenge when hosts are densely packed with VMs is that the head room that IT used to count on to cover unexpected peak load situations is eliminated. It also puts a special burden on storage I/O since each of the VMs now generates I/O requests and this I/O is generated randomly without coordination with the other VMs. To make sure these potential storage challenges don’t negatively impact overall infrastructure performance, planning is required prior to moving to a densely packed VM environment.

Understanding The VM I/O Requirements

There are, generally speaking, three types of VMs and the I/O requirements of each of these VMs must be well understood for densely packed VM architectures to function. One of the first goals of densely packing VMs is to move existing VMs from older, less powerful hosts and centralize them onto a fewer more powerful hosts. Retirement of these older hosts will represent a reduction in the expense of powering and cooling these systems, since typically the older a system is, the less energy efficient it is. Removal of these systems will also free up infrastructure like switch ports, cabling and even potentially provide extra network interface cards. Finally it will also eliminate any maintenance expense associated with those hosts.

These VMs though must first be analyzed so that their I/O profile can be understood. While manual utilities exist from VMware that can provide this information, typically these tools can only examine one VM at a time. That means that all the data from all the VMs needs to be manually correlated. Also the built-in utilities typically report only the I/O load of that VM at the point in time that the analysis is run. They may entirely miss peak loads.

To capture this information across all the VMs of a host or hosts at the same time requires a third party application, not a utility. Companies like SolarWinds provide software that can analyze the current VM storage I/O workload, automatically correlate those VMs into a single view and then trend that information over time. This provides the administrator with the tools they need to be able to determine which VMs and how many VMs can be moved to fewer hosts, allowing them to achieve greater density.

The movement of existing VMs to fewer hosts will, as discussed above, reduce power, cooling and infrastructure costs. It is also, with the proper tools, the component that is the easiest from which to get very accurate I/O profiles since they are already in the virtual mode.

Understanding Stand-Alone Application Requirements

The second opportunity to increase VM density per physical host is to examine the servers that have not yet been virtualized to determine potential candidates for virtualization. It is important that these stand-alone servers do not become the motivation for a new host but that instead they are integrated into the existing infrastructure.

To safely make the consolidation requires two pieces of information. First is the available I/O profile of the physical hosts and second, the requirements of the stand-alone application servers. Once again applications like SolarWinds Virtualization Manager can monitor I/O on the virtual host while Storage Manager provided the impact on the underlying array. They can trend over time the total I/O consumption of a physical host and they can trend over time the I/O consumption of an individual application server. With this information the right applications can be virtualized and be placed on the hosts that have sufficient I/O capacity.

Understanding A New VM’s Requirements

Finally as the data center adopts a "virtualize first" strategy, it is important to determine the I/O needs of the newly requested VMs. The problem is that there is no active workload to examine and often the I/O demand is determined from comparing similar workloads.

Instead of going with a best guess strategy it may be better to have a host designed specifically for new VMs where the VM count is purposely small and resources are purposely inefficiently used. This allows the new VM to run in the virtual infrastructure so it can be analyzed and an I/O profile built using the tools described above. Once that information is captured then it can be reintegrated into a more densely packed virtual environment.

Understanding the Densely Packed VM Architecture

While hosts densely packed with VMs represent an opportunity to extract additional ROI out of the virtual infrastructure as well as a method for delaying future expenditures they also confront IT with a key challenge. The headroom that used to be counted on in case of an unexpected spike in I/O demand may no longer be available.

The lack of this spare I/O space means that the environment needs to be actively monitored and analyzed for any current or upcoming fluctuations in workload. If this analysis is done by a human it will negate the ROI gained by the density project in the first place. If this analysis is done by software that only captures information at certain intervals it may miss the information needed to correctly predict spikes or warn that one is occurring.

Instead this monitoring and analysis must be done continuously by software in real time so that the moment there is a change in usage, corrective action can be made by an administrator. This allows the administrator to focus on other duties while the software "stares" at the environment. It also provides more accurate trending information since there are no gaps in what is being recorded.

Conclusion

Hosts densely packed with VMs are the only way that data centers can realize the full potential of their virtual investments. However, to make sure that the densely packed virtual environment does not cause more harm than good requires that the IT professional be armed with the tools required to make accurate decisions on VM placement and to be able to monitor the impact of those decisions. Software applications like those from SolarWinds provide the ability to achieve high VM density hosts safely so that maximum ROI can be realized.

SolarWinds is a client of Storage Switzerland

Previous Entry: “The Cloud Storage Provisioning Roadblock”