Cost Justifying SSD In The Virtual Environment

While the potential performance upgrade of SSDs in a virtual environment may seem somewhat obvious there is bound to be a concern about upfront cost. In the second article of this series Storage Switzerland will discuss the performance improvements that SSD brings to both the server and desktop virtual environments. First, though, virtualization must be cost justified. SSD is not typically thought of for a ‘horizontal’ application like server virtualization. Traditionally solid state storage has been relegated to niche use cases where extreme performance can warrant the additional cost by directly correlating the performance boost to increased revenue. By enhancing server and desktop virtualization with SSD, improved performance may drive out costs in the virtual environment and may be the perfect horizontal use case for SSD.

Reducing The Number Of Hosts

First, the added performance capabilities of SSD allows for denser VM or virtual desktop populations and reduces the number of physical servers needed. Keep in mind that physical servers in a virtual environment are usually more expensive than a typical stand alone server. They are often loaded up with additional memory as well as network and storage I/O. The fewer hosts that are needed in the environment, the lower the server virtualization budget can be. It should not go unmentioned that fewer physical hosts can also lead to the reduction in server virtualization licensing costs since most of these software applications are licensed either by host or by processor within the host, both of which are almost always reduced by using solid state storage.

Host Cost Reduction

SSD can possibly even help to lower the cost of the hosts that are needed. One of the biggest expenses in host servers is memory. The cost of fully equipping a server with memory can be exorbitant and in certain environments, even a server with the maximum amount of memory installed won’t be sufficient to support the VMs on that host. As a result server virtualization environments have a constant need for swap files. The objective behind adding more RAM to a server is to reduce the amount of swap file activity that has to occur. In an SSD-enhanced virtualized environment the speed at which those swap files can be accessed and utilized is almost as fast as the DRAM in the server itself. As a result the amount of memory per server may be reduced significantly.

Enhancing Capacity Optimization

Another area that will help reduce the cost of SSD in the virtual environment is with one of the many space optimization products that are available. Technologies like snapshots, clones (writable snapshots), compression and deduplication are becoming commonplace in the storage systems that support the virtual environment. By leveraging technology that supports these capabilities a virtual environment can commonly see 20:1 space optimization. Normally the space efficiency gains come at the cost of degraded performance. Again, here’s where solid state shines. First, from a raw performance perspective, SSDs have IOPs to spare in most cases, so using a little performance to gain this type of optimization on premium storage is justified. Second, much of the I/O impact that storage optimization technologies cause is read-related, something that SSDs are the strongest at. Overall, the payoff on optimization can be particularly compelling with SSD.

Leveraging Machine Migration

The net effect of optimization is that there’s the potential to store the entire image of the most active VMs on SSD and do so cost effectively, by keeping SSD capacity relatively small. Hypervisors like VMware’s vSphere include the capability to quickly move a VM between different storage devices in real time, while the VM is active. Essentially VMs can be promoted or demoted between storage platforms and storage protocols on the fly, without user interruption.

For example NAS is often thought of as the ideal, low cost, simple way to support hundreds of VMs. The challenge is that the Ethernet segment and the NAS itself can  be potential bottlenecks. With SSD and Storage vMotion the environment can be monitored for the VMs that are generating the most storage I/O and those VMs can be moved back and forth between a fibre attached SSD and NAS as the I/O profiles change.

Tuning Time

Another aspect of enhancing server virtualization with SSD is the management time saved by not continually fine tuning the environment. When storage I/O becomes a bottleneck there are a wide range of options available to tune performance for specific hosts or even individual VMs. Some of the options involve setting priorities on performance hungry VMs, dedicating bandwidth to specific VMs or expanding the overall I/O of the host. They also may involve redesigning the storage array layout so that it’s better suited to the highly random nature of the virtual infrastructure’s I/O demands.

While tuning options are always important they all require time, knowledge and skill in implementation. This is especially true in the abstract and rapidly changing virtual infrastructure, and many of these tuning methods often induce risk as well. SSD for the most part removes much of the need for tuning. It simply solves the problem by almost always providing more performance than the environment will need or at least outperform the most highly tuned mechanical based systems.

SSD Density

An important aspect of SSD is the method that’s used to deliver the systems. One of the biggest cost challenges that traditional storage system suppliers face in delivering SSD to virtual environments is their choice of solid state drives instead of solid state devices or systems. Solid State Drives are memory based storage systems that are packaged in a traditional mechanical drive form factor. While these are fine for use in single server implementations they lead to wasted space in legacy storage arrays. This is because legacy storage arrays were designed for mechanical drives and need to account for the side effects of running those drives. For example a spinning drive creates significant heat, vibration and harmonic issues. These all need to be compensated for by providing more space between the drives and limiting the number of drives per chassis or drive shelf.

Since SSDs need to go into these same shelves the manufacturer cannot take advantage of the fact that solid state technology doesn’t suffer from those issues and more devices could be packed into a single shelf. Additionally, the shelves that hold traditional disk drives are designed to the performance specifications of mechanical drives, not solid state storage. Storage processors and arbitrary loops quickly become bottlenecks when implementing SSDs into traditional hard drives arrays. As a result many manufacturers do not recommend filling up an entire shelf with SSD, the preference being to leave the majority of the shelf empty, which leads to more wasted space and increased cost.

Solid state-specific devices or appliances like those from Texas Memory Systems are designed exactly for solid state storage. They allow for very dense packaging of solid state storage modules and have the I/O bandwidth designed to handle the maximum capabilities of memory based storage. This density means that fewer shelves, power supplies and I/O channels need to be purchased per TB of storage and leads to significantly reduced prices versus the traditional storage manufacturers.

SSD Integration

These solid state devices can be easily integrated into a virtualized server or desktop storage infrastructure. There is a choice of allowing them to be part of an existing storage software solution and have that application manage the device or to leverage the tools within the virtualization product to offer the expected data services to the device. We will cover the integration options in greater detail in the next part in Storage Switzerland’s series, Enhancing Server and Desktop Virtualization with SSD.

Texas Memory Systems is a client of Storage Switzerland