In 2013, the VMware space has seen a lot of innovation as customers continue to virtualize more and more of their data center workloads. The goal for many is to virtualize as much as possible and look to do it in a more economical way. These two goals are driving innovation from VMware and their partners. They are looking for news to provide performance, cost savings or both in effort to allow customers to continue down this path. One of the ideas in this discussion is finding new ways to utilize direct-attached storage (DAS).
The directed-attached storage market has been around since the beginning of servers. The first place that storage was available was within the computer itself. In today’s market DAS can be either storage that is directly inside of the server in the form of a drive that is in a bay of the server or on a PCI card inside the server. The other DAS option would be storage that is located in a separate enclosure that is directly connected to the server via a SAS or SCSI connection. Both of these methods have allowed for the local storage capacity to be greatly increased.
The modern servers today have many models that can offer upwards of 20 local drive bays. These servers allow for a large amount of capacity to located within the server. But depending on the types of storage purchased you may still be limited on the performance based on the number or types of disks that are purchased.
Now that we have a basic idea of what DAS is, let’s talk about how we might use it. In years past, DAS was used on a per-server basis. This meant that a single physical server would use the local storage for its needed capacity and performance required for the workload. The limitation with that method is that you were bound by the performance of the number and type of disks that you had in your server. There also was no caching.
Today there are a number of companies innovating in this space and are using DAS in some creative ways. There are really only two common methods that these companies are focused on. The first method uses local storage as a caching layer, typically a small amount of flash storage. The other method takes all the DAS storage from a cluster of servers and unifies it into a single storage layer and adds a caching method.
This is the exciting part of this discussion. With the ability to take a bunch of regular spinning disks and some flash disks, and to locate them in servers that can be collapsed into a single pool of storage is pretty exciting. This allows us to use the capacity from the spinning disks and take advantage of the performance of the flash storage to meet the performance demands that workloads need.
This method is allowing customers to bring the storage closer to the applications and provide the same or better performance that they were accustomed to receiving from their shared storage systems. By using flash storage in each server, these vendors are able to create some very high-performing storage without the cost or increased complexity to traditional storage arrays.
Okay, now that I have painted the picture about what is possible, I wanted to briefly list out a few companies that are working on this technology. Do not consider this a validation or recommendation of any of them – these are simply a few vendors that I think are exciting and have different approaches to this technology space. I encourage you to learn more about them.
VMware – No surprise here, right? The company that brings you the best hypervisor on the planet continues to look for more ways to increase adoption. VMware announced VSAN back at VMworld this year. VSAN is still in public beta and is expected to be released first half of 2014. VMware is taking a combination of spinning disks and flash and creating a shared datastore within a vSphere cluster. The spinning disks provide the capacity, and the flash storage is there for caching to increase performance. Early feedback is the product is performing well, but we still do not know how VSAN will be sold. I think the real success of VSAN will depend on whether VMware includes it with some licensing levels or decides to license it separately. If it’s sold separately, the cost must be in the right range or it will just get ignored.
Maxta – This is a brand new company that I recently heard about. I think of it as VMware VSAN but with more features and its not tied to VMware. Maxta will allow you to run its local storage technology on hypervisors from multiple vendors. If you have a environment that supports more than one vendor for virtualization, this could be attractive to you. For those shops adopting Hyper-V or KVM as their platform, then Maxta might be worth a look.
Hyper-Convergence – This approach to local storage is being tackled by more than one vendor. The hyper-convergence idea is packaging compute nodes with local storage and a storage layer all in one product. So this model requires you to purchase the whole solution rather than just a storage offering and then use your servers and disks. I guess the value pitch to this discussion is these vendors are tuning their offerings to work perfectly on the hardware platforms that they are building. This should allow them to offer and guarantee performance easier. The two vendors that I consider to be leading in this space for VMware shops would be Nutanix and Simplivity. I have not used either of these platforms, but I have been briefed and the solutions are very interesting.
I think that there are some great solutions being worked on in this space, and I’m looking forward to seeing these vendors and others mature their solutions. I also tend to think about how I might use these types of solutions for different projects and designs. It’s good to think about how they might help with costs, speed of delivering the solution, or increasing performance. If you have used any of these options, or others drop us a note in the comments section – it’s always good to hear how things worked for customers.