One of the reasons that server virtualization has been so widely adopted is that its return on investment (ROI) is easy to calculate. If you replace 150 servers with 10 hosts running 15 virtual machines each you can add up the cost savings of not having to power, cool and manage 140 servers. That’s a significant savings.

After the initial implementation is complete, each host you add to the virtual environment lowers the ROI. As a result the goal should be to increase virtual machine density as much as possible on each existing physical host before you add more hosts. Other than available RAM in the server, storage is a primary culprit in limiting how densely packed a host can be with virtual machines.

There are essentially four choke points that need to be opened up so that storage I/O can flow and virtual machine density can be increased, and these can be addressed with a change to the storage infrastructure. The first is the RAM issue. While it’s not exactly a storage problem, a storage solution may help. After all the first step that the hypervisor will take when it runs out of RAM is to start paging older memory segments out to disk.

As we described in our article “What is Storage Memory” a host that is densely packed with VMs to the point that it’s paging to disk for extra RAM can benefit greatly by using an SSD device. This is the ideal use case for PCIe SSDs since they have high speed access to the CPU and in this type of operation data doesn’t need to be shared outside the server.

The second choke point is the storage network itself. As each host is packed down with more virtual machines it becomes an I/O nightmare for the storage infrastructure. The good news is that upgrades to the speed of the storage network are readily available. 10GbE is becoming more affordable than ever and 8Gb/16Gb fibre channel are seeing similar price decreases.

Raw speed is great but at some point controls will need to be in place to make sure that mission critical virtual machines are not roadblocked. Cards are available now that let you prioritize workloads on the storage network, so that critical virtual machines always have a priority level of bandwidth over other VMs. This capability is critical for administrators looking to virtualize more of the companies’ business application servers, but concerned about meeting SLAs.

The third choke point is the storage controller itself, as we have written about in InformationWeek columns. The amount of I/O that each host, now loaded with VMs, must deliver across that high speed network is daunting. This is where high performance controllers or scale out storage designs are needed so this inbound storage traffic can be received.

The final choke point is the disk spindle itself. As we discussed in our article “How Flash Storage Can Impact Server Virtualization”, solid state storage is becoming a key element in addressing this last point of contention. The options for solid state systems are endless, from all-flash solutions to those that mix flash with mechanical hard drives. Which one to choose really depends on the environment and how dense you want to pack virtual machines into a cluster.

Increased VM density is key to ROI, and the next step once the first wave of virtualization is complete. However, that density puts a tremendous load on the storage infrastructure. The good news is that there are storage options available that allow you to eliminate it as a roadblock. The next big challenge is protecting these densely packed VM environments and we will cover that in an upcoming column.

previous entry: “Vertical Integration an Advantage for SSD Suppliers”