Over the next few columns, we will continue our discussion on data storage systems and look at how they are evolving to cater to the world of data-centric computing.
Last month, we had started our discussion of storage systems by looking at various concepts like SAN, NAS, etc. In this column, we will look at the concept of scale-up vs scale-out storage, and discuss their relative advantages and disadvantages. up storage system, you add more and more storage capacity to the single storage node to meet increasing storage requirements. This can be achieved by adding many individual disk drives to a storage controller (or a pair of storage controllers so as to ensure failover and high availability). Since the storage capacity ( i. e., number of individual disk drives) that a storage controller can support is limited, if storage requirements exceed that capacity, the only option is to move to the next bigger controller, with a higher capacity. With scale-out storage, this problem is avoided, since each node has its own storage controller, and as you add more nodes to meet increasing storage demands, it allows the controller architecture to grow as well. Scale-out storage is typically marketed under the slogan, “Pay as you grow”, since it allows you to add nodes as and when your storage requirements increase, instead of having to over-provision from the beginning itself.
The next question that arises is about performance and cost. How do scale-up and scale-out storage systems compare in terms of performance? Before that, we need to understand what the common measures of performance for storage systems are. For computing systems, you can specify its performance in terms of its clock frequency (1/clock frequency gives the clock cycle period) and Cycles Per Instruction (CPI). It is possible to approximate the execution time for a program as (clock period G CPI G Instruction Count). For a computation system, the number of operations/ instructions completed per second is a performance PHDsXUH. TKLs Ls WySLFDOOy sSHFLfiHG Ds KRw PDny MIPS (million instructions per second) or how many