Many predictions have been made about if and when Solid State Disk (SSD) will become the dominant storage platform in the enterprise, especially in primary storage. For a particular type of workload that time may well be upon us. Solid State Disks are cost effective now for data sets that are more performance sensitive than they are capacity sensitive.

SSDs are ready to move out of the "Performance at any Cost" selection criteria and into merely performance being more important than capacity. There are plenty of important, not just mission critical, workloads where the required capacity for the lifetime of the application will not exceed 500GB, less than the size of today's average single drive capacity. By comparison PCI-e based Solid State Disks like those from Texas Memory Systems can deliver 400+ GB's of storage on a card for less than $18,000.

While certainly this cost is more than the cost of the comparative capacity single mechanical drive, typically for a mission critical or even merely important workload the data is not going to be stored on a single drive because of reliability and performance issues. They will likely be stored on an internal or external RAID with mechanical drives. To get the performance and redundancy, the capacity of the direct attached system is far more than a typical workload of this type requires.

Compare the complexity of configuring, powering, and maintaining a RAID plus the wasted capacity, to the simplicity of installing an SSD PCI-e card with no moving parts, built-in RAID protection, reduced power consumption, and orders of magnitude higher performance. All with the exact capacity that the workload will require.

In a shared storage environment one of the major justifications of networked storage is the leveraging of capacity across multiple servers and their workloads. The challenge in a shared workload becomes one of balancing maximum capacity utilization vs. optimal performance. As a result, large array sets are created to improve performance but are cost justified by placing more workloads on those array sets. The result is that the often manual task of load balancing storage IO is laid at the feet of the already burdened storage administrator. This administrator must try to match high IO requirement workloads with more modest workloads to determine how to effectively use the available capacity while not negatively impacting the performance of performance sensitive workloads.

This challenge leads the storage administrator to create different RAID groups for different applications and then create LUN's underneath those array groups for assignment to the various servers and workloads. Each array group will house only a few performance sensitive workloads, each with LUN's assigned to them. The rest of the servers assigned to each array group will be less performance sensitive. In today's data center this is then compounded by server virtualization; now a single server will have multiple workloads - each with differing performance needs that are significantly harder to isolate.

While shared storage was designed to optimize capacity utilization, it was also created to provide a large number of spindles to meet specific performance requirements.  This creates a complex environment where, ironically, the array groups designed to isolate performance do not effectively use capacity. Most of the workloads that need this storage IO performance have smaller capacity requirements compared to today’s SAN capacities.

Complexity and capacity utilization are not the only issues; there is a hard cost issue as well. While the capacity per mechanical drive has been increasing dramatically over the past few years, the cost per drive has been relatively stagnant. In both the SATA and Fibre Channel cases the cost per mechanical drive has stayed relatively stagnant and the performance per drive has absolutely been stagnant; 15K RPM drives have been the maximum performing drives for years.

The problem of capacity per drive increasing while the cost and performance of each drive remains flat becomes particularly important when the workload moves from performance sensitive to performance critical. To overcome the per-drive performance limitations, customers will create array groups with very high drive counts and in some cases use short stroking to ensure that the data is placed on the outer edge of the platter. All of this leads to even greater complexity, greater capacity underutilization, increased power and cooling costs, and of course even greater purchase costs.

As a result of this paradox it may be simpler and more cost effective to move the workloads that have smaller capacity requirements but higher performance requirements to dedicated SSD based systems. While it is true that this is potentially a separate storage array to manage, it is not much different from managing multiple array groups in a mechanical drive based storage system and can simplify the array group management on the SAN.

By moving modest capacity and performance friendly workloads to a dedicated SSD and then keeping the capacity friendly workloads on a smaller number of array groups the management complexity of the storage platform is greatly reduced while the user experience of workloads where performance matters is improved.

With this design strategy not only is the cost associated with managing storage greatly reduced but also the capital acquisition costs are actually reduced as well. With the performance sensitive applications moved to SSD, fewer array groups must be created. As a result more servers and workloads can be deployed on these fewer groups, resulting in more effective capacity utilization and fewer physical drive purchases.

In many cases a single SSD system can accommodate all the performance sensitive and performance critical workloads. This will certainly require less power, cooling, and space than comparable arrays, with 30 times the performance of even the highest performing mechanical drive based array. After the costs of maximizing the performance of mechanical based disk array systems, hard dollar cost comparisons begin to tip in favor of SSDs. The result, for an increasing number of customers, is Solid State Disks are Cost Effective NOW.