VMware changed the data center so much that it is hard to imagine a world without virtualization. One challenge for VMware customers (still) – as well as customers of other hypervisors – is storage technology hasn’t always kept pace with the demands of virtualization. Nowhere is this more evident than in VMware’s decision to create its own storage product with VSAN. Yet even this product, which was designed with VMware in mind, comes with its own limitations. So what should one expect from VMware storage? Before answering that question, let’s take a look at the advantages and disadvantages of the various storage options for a VMware environment.
Scale Up Storage Arrays
The first storage option for VMware is the traditional scale up storage array. This is the storage product that dominated the market when VMware first came on the scene, and it continues to dominate today. Scale up storage arrays are still around for more reasons than inertia. Scale up technology is well understood. You purchase a storage array of a certain size, and if you need to make it larger, you just buy more shelves. Everyone understands that each storage array can only scale up so far, but this is all taken into consideration during the initial purchase.
The scale up storage challenge is that these arrays can only scale as far as the control head purchased with the array will allow. Those purchasing a scale up storage array must decide in advance how far the array will eventually need to scale, and purchase the appropriate control head to go with that scaling expectation. Human nature tends to cause them to buy larger control heads than they otherwise would so they do not end up replacing the control head and wasting money.
Traditional scale up storage arrays also tend to be more expensive because they are often built on purpose-built hardware with custom ASICs that are not mass produced. The custom hardware increases the cost per unit for many reasons, starting with the fact that the unit actually costs more to produce. Having the unit based on custom hardware affords them more control over the quality, but mostly gives the vendor the opportunity to increase its margin. The customer cannot get the functionality provided by the storage system without purchasing the hardware from the company producing the array.
Another thing that tends to increase the cost of traditional scale up storage arrays occurs when one chooses to use replication to provide off-site data protection. Almost all scale up storage arrays can only replicate data to an array of the exact same type. The only thing more expensive than a scale up storage array is two of them.
Scale Out Storage Arrays
Scale out arrays were designed to address the previously mentioned scale up array challenges. Almost all of them are built on commodity hardware using the same Intel chipset and RAM chips that go into every server. Using mass-produced hardware drives down the cost of producing the units and makes it harder for vendors of such products to increase their margin simply by charging more for the hardware. The margin tends to come exclusively from the software that turns an ordinary server into a storage array.
Scale out storage arrays are much easier to scale without prior planning. That is, unlike a scale up storage array, it is not necessary to understand how large a storage system needs to grow prior to purchasing the first few nodes. You can start your storage system with enough nodes to meet your performance, data protection and availability requirements, and add to those nodes as your needs’ dictate. There is no need to buy more expensive nodes in the beginning like there is with scale up storage arrays.
Replication between arrays in a typical scale out storage array is between like versions of software, not like versions of hardware. With a few exceptions, the storage operating system does not care if the data is written to an SSD, Fibre Channel disk, SAS disk, or SATA disk. This allows customers to use more expensive disks for the primary storage array and less expensive disks for the secondary copy located off-site. Using different classes of storage to accomplish different purposes lowers the overall cost when compared to traditional scale up storage arrays that must be replicated between like hardware.
Challenges with both types of Arrays
Both scale up and scale out storage arrays still have a number of challenges they share. The first challenge is neither option really allows you to scale down. Once you purchase additional storage capacity, you are stuck with it. It may be easy to scale up in either choice, but getting rid of hardware once it is been purchased is near impossible. It is true that hardware you purchase for a scale out storage array could be repurposed, but the value added software that came with it is often capacity based and cannot be repurposed.
The other downside that scale up and scale out storage arrays share is the length of the typical purchase cycle in a medium to large company. It can take many weeks or months to authorize the purchase of additional capacity. If the cost of the additional capacity was not included in this year’s capital budget, the purchase could be delayed until the following year. Exceptions can be made of course, but they require many levels of approval that most IT managers would rather not undertake. Therefore, human nature dictates that people purchasing scale up our scale out storage arrays tend to purchase significantly more storage than they need in order to deal with this lengthy purchase cycle.
The final challenge facing scale up and scale out storage arrays is that very few of them understand that cloud storage exists. Such devices could, for example, use the cloud as a tier. What does that mean? It means a full copy of the data is moved from the array to the cloud. If you’re a backup person, think of this as the cloud version of “tape out”. This provides the advantage of storing data offsite and potentially at lower costs, but it’s not cloud bursting and doesn’t take advantage of the agility clouds can provide. Simply put, scale up and scale out array vendors don’t think about the cloud, and one can make a strong argument that it is against their best interest to do so.
VMware got tired of many of these same limitations and decided to design their own storage; VSAN was the result. Like scale out storage arrays, VSAN uses commodity hardware. In fact, it uses the same hardware that is already being used for VMware. It is very easy to use in a VMware environment, and to a certain extent it helps solve the purchase cycle problem. It is much easier to purchase a few additional disk drives to scale VSAN than it is to buy an entire shelf of storage for a scale up array or an additional node for a scale out array. It’s even possible that the purchase of additional capacity could fall under expense guidelines rather than capital purchase guidelines, decreasing the length of the purchase cycle in larger companies.
VSAN has its own limitations, though. It is able to replicate between sites, but only between two sites. It recently added iSCSI, but really isn’t meant to be used as an external storage device. It only scales to 64 nodes. Although there are many environments that will never hit this number, larger enterprises certainly will. Like most scale up and scale out storage arrays, it also has no knowledge of the cloud. One of the biggest limitations, though, is that VSAN is a VMware only product. If you change your mind about what hypervisor you’re going to use, or would like to use more than one hypervisor, VSAN is of no help.
Is The Cloud an Option?
What about the cloud, then? Is it possible to use cloud storage for VMware? Cloud has the ability to scale up and down, and purchases fall under expense guidelines. Customers can immediately scale up their storage system if need be, and – unlike any other storage system – scale down their system once their needs change. There’s only one problem: you can’t use cloud storage with VMware. It is true that VMware and Amazon are partnering and you will soon be able to run VMware workloads in the cloud. But you will still not be able to run a VMware workload in your data center on storage in the cloud. Therefore, without special design considerations, VMware in your data center and storage in the public cloud is a nonstarter.
So what should you expect from your VMware Storage?
The next wave of storage systems are those built using software-defined and distributed techniques. A modern, software-defined storage system should have the advantages mentioned above without the disadvantages. The design should be based on commodity hardware. Doing so makes sense and reduce costs all around. It should be easy to use with VMware, while not completely tying you into VMware. If you decide to run multiple hypervisors, or if you decide that you’d rather run a completely different hypervisor as your main hypervisor, your storage should work with it as well. It also should support bare metal systems, as most data centers have one or two critical applications that may never be virtualized. Obviously the storage should be able to easily scale up and down to meet your needs. Being able to scale up is easily done with a scale out storage architecture, but being able to scale down is really only possible by integrating the cloud for periods of peak demand.
The storage system should be able to understand more than just two-site protection. It should be able to distribute storage across locations and the cloud but still be managed as a single cluster. As part of that data distribution it should allow you to define how many simultaneous outages you are willing to tolerate. If Amazon’s storage system can survive multiple simultaneous outages, then so should the storage system you use in your data center.
Storage has a long way to go for use in a VMware environment. For smaller environments, VSAN and other solutions are fine. You probably don’t need the scaling and cloud-bursting features, given the smaller, more predictable nature of these environments. But for larger environments, scale up, scale out, and VSAN aren’t always a great fit. They cost too much and are too hard to scale up and impossible to scale down. The cloud can help but only if it’s correctly integrated and not just bolted on as a cold storage tier. Modern, software-defined, and VMware-friendly storage offers you these advantages, plus affords you flexibility to add other hypervisors and workloads.
Sponsored by Hedvig