Originating Author: Luca Bert
One of the most identifiable trends in the storage arena in recent years has been the explosion of RAID solutions in all server market segments. This has spurred an incredible amount of new solutions and their gradual transitions toward external storage.
There are multiple reasons for this transition, but the key advantages of external RAID are: - Better availability across multiple platforms (failover) - Better provisioning - Better manageability - Better TCO (for larger configurations at least)
On the other side, direct-attached storage (DAS) always had (and still has) some major advantage in the cost structure that makes it more suitable for very small configurations. This includes both the capital cost of equipment and the management costs as the skill-set required to manage it.
However, looking forward, these lines are getting blurred as new technologies, most of all server virtualization, are starting to become more prevalent. Soon people will be looking at server and storage platforms in their entirety rather than as independent platforms. And DAS RAID may be one of the key engines that enable this ubiquity.
When looking from a user point of view at the viability of DAS RAID in the small-to-medium sized business (SMB) environment, the question becomes: “How small is a small business where DAS makes sense?” For example data centers are definitely focused on SAN architectures: FC historically, now moving on fringes like SAS, FCoE, 10GbE iSCSI, etc. They can realize the highest capex and TCO optimization for them as driven by the sheer volume and storage capacity they need to manage.
However, smaller customers may not see all those TCO benefits while still paying for the capital expenses associated with SAN. How much lower in the configuration arena does one have to go before finding people more comfortable using servers with internal DAS? Probably not much; SMB is a market where the “S” part of customers are most likely DAS users, while the “M” customers are much more amenable to intermediate solutions that can bring in at least part of SAN advantages, assuming the cost is not prohibitive. So these new markets are ideal for 1 GbE iSCSI and SAS-based solutions.
In the future there are new technologies coming that may lower the cost (capex and opex) of operating in SAN-like configurations, generally referred to as “Virtual SAN”. The concept is that a SAN is a very well defined concept where a number of computing nodes (servers) share a number of storage nodes (SAN RAID boxes), but this physical description can also be applied to virtual configurations that are not SAN in physical implementation (they are standard servers), but that run services that expose storage as if it were on a SAN, making it a Virtual SAN configuration.
In this fashion, imagine a server farm, where each server has only DAS elements (i.e., optimize server capex), which are exposed to all other nodes using standard, though dedicated network capabilities that are provisioned as needed (thus optimizing solution capex), and managed through the usual provisioning methods in use in conventional SAN topologies, while providing the same and better RAS, granular scalability, etc. (thus optimizing opex as well).
One of the key ingredients to make this viable is server virtualization, i.e. the technology that allows multiple operating systems to run on a shared server platform as Virtual Machines (VM), and creates opportunities for having some of these VMs to become the engines for virtualizing the local storage across the network while supporting all the critical features, including the VM mobility that is currently a major limiting factor for DAS solution on virtualized servers
While this solution is probably not threatening the data center applications since they already have a stable model that is addressing the TCO requirement in an acceptable (by most) way, these solutions will be much more suitable for the smaller customer, both “S” and “M” customers in the SMB segment, and a large class of users.
In this deployment model, only standard servers are used with a software infrastructure to create a Virtual SAN. The storage controller of each node is a conventional DAS RAID that becomes a building block, Lego style, of a much bigger storage network.
It is also noticeable that while any DAS RAID controller meets the minimum requirements to create such infrastructure, the tight integration of the DAS RAID controller, with the server node (and the hypervisor) and the storage virtualization software, offer a class of solutions and services, a scalability and a performance level, that are well above SAN in a comparable price range.
Action Item: One needs to separate the cause and the effect of any major trend. In case of storage, the current transition from DAS to External Storage is the “effect” due to usage model needs and not the “cause” driving it. Other models and other technologies that may allow serving the same customer needs may reverse this transition until a more technical and causal balance between them is reached.
Footnotes: