Scale up storage typically means a storage system with two controllers that are responsible for all server or front side I/O and for all disk or backside I/O. There is a concern that these controllers can become overwhelmed as the potential front-side bandwidth increases, the number of backside disks increase, and the number of services (thin provisioning, snapshots, automated tiering) they provide increases.
This concern has lead a number of storage vendors to introduce scale out storage systems. Systems based on a cluster or grid of servers, often called nodes. Each time a node is added it adds to the aggregate capacity and performance of the scale out storage system. The idea being that as your environment grows the performance of your storage system should scale linearly with it.
Scale up storage is not dead, though. As we discuss in our article "The Advantage Of Scale Up Storage," there are many ways to scale a scale up storage system. First of all, the capabilities of the initial unit have a wide range of scaling--many initial systems can start out quite small and scale very large. Second, there are software additions that allow additional independent scale out storage systems to be added to the environment but still keep it manageable. The most common is a graphical user interface that allows the management of multiple systems from a single interface. The newer capability is the live volume migration that I covered last entry.
Live volume migration, either from the storage vendor or built into a hypervisor like VMware, allows you to transparently move a volume from one controller to another without interrupting that application. If the second system is not as busy as the first or if it is more powerful, the application should have access to better storage performance.
This does not make live migration a scale out storage offering though. You are still limited to the performance potential of the controller pair that the volume ends up residing on. It does not provide an aggregate performance like an array-based scale up storage system does. That said, for many data centers, two or three of these scale up type of architectures armed with a live volume migration capability is going to provide more than enough performance for their applications.
As you can see in many cases, there is no real distinct advantage between the two architectures (scale up vs. scale out). There are different points in the growth curve where one will outperform the other, especially when price is factored in. As I have suggested in the past, our recommendation is to make a performance decision based on how much performance can you get that will meet your capacity needs and stay within your budget reality. For example, if you have $100,000 and need 10 TBs of storage, how much performance can you get for that amount from each vendor?
There are other factors in the scale out vs. scale up decision. Ease of implementation, management, and expansion, for example. Another is performance at capacity. One advantage that you would assume that scale out storage would have is the ability to maintain steady performance no matter how large your data set grows. We'll explore that in more detail in our next entry.
Follow Storage Switzerland on Twitter