Storage: In Scale-up Versus Scale-out Wars, Scale-out has Momentum

A recent article on GigaOm compared the storage strategies being promoted by HP and Oracle.  The article by Barb Darrow, senior writer at GigaOm,  commented that the difference between the two approaches was a perfect example of “a classic scale-up vs. scale-out battle” of storage strategies.  What exactly did she mean by that?

Scale-out, sometimes also called horizontal scaling, refers to expanding system capacity by adding more hardware units or nodes.  Scale-out strategies often try to expand capacity by combining large numbers of cheap commodity hardware units.  For example, hundreds of low-end commodity computers could be combined into a cluster which provides computing power which equals or exceeds that of a single supercomputer.  The techniques relies on the availability of very fast interconnected communication between the individual nodes.

Scale-up, sometimes also called vertical scaling,  refers to expanding system capacity by bulking up the resources of a single node or machine.  It might involve adding more RAM or CPUs to increase performance.  Or more commonly, it would mean upgrading the entire hardware unit on a node with a new upgraded and more performant piece of hardware.

Scale-up strategies tend to be simple, while scale-out ones are often complex.  Although new scale-out technologies are trying to close that gap.

A big disadvantage of scale-out, for example,  is that as more nodes are added to a system, management of the many nodes can become very complex and there is a potential for throughput and latency issues.  Often programs or applications also need to be re-written or re-structured to take full advantage of the scale-out architecture.  On the other hand, the big advantage of scale-out typically is the cost — it tends to be a significantly more cost effective approach.

Storage is one area where scale-out architectures are being increasingly applied.  This trend is being driven by challenges IT shops are faced with in managing rapidly growing amounts of data.  Large organizations are increasingly going down the path of scale-out, adopting network-attached storage (NAS) systems, an approach that allows them to fairly easily scale to multi-petabytes of  data.  Smaller companies are testing elastically scalable cloud-based storage solutions, most of which are implemented with scale-out storage technologies.

In commenting about scale-out NAS (network-attached storage) storage technology, Terri McClure, a senor analyst at ESG, said that “scale-out NAS really started in niche markets roughly a decade ago, so it’s certainly not brand new. These systems typically excelled in handling high-throughput applications such as media, entertainment, geospatial imaging or bioinformatics applications, and that’s because of the bandwidth they can provide. But what is new is that we’ve seen a class of products emerge that fits the more file I/O-intensive performance requirements typical of enterprise IT shops.”

But new NAS scale-out technology can not be considered fully baked just yet.  Unlike more mature scale-up storage technologies, scale-out technology doesn’t yet have the same wide range of existing add-on products for handling operations like snapshots, one-to-many and many-to-one replication, remote replication and remote snapshots, thin provisioning, deduplication and compression.  Although it is likely only a matter of time.

Arun Taneja, founder and consulting analyst at Taneja Group in Hopkinton, Mass, said that there is a current limit to NAS scalability.  He said that “you can put them [more storage units] together and they’ll work. But you will have a declining performance curve.  The curve will be pretty linear for a while, and then at some point in time, it’ll start flattening and, further out, it’s going to actually start going down.”