The Storage Economist

With the increase in new storage architectures, there are new metrics that can help show continuous improvement in cost, scale and purchasing leverage. For years we have used a $/GB metric to compare and contrast purchase costs of storage. Thanks to Moore’s law, this ratio keep getting better for the buyer each and every year.

When we talk about Total cost, we often use a Unit Cost metric of $/TB/year, or something similar. This metric looks way beyond the purchase price and should include other operational and capital costs such as:

• Labor costs for storage admin, backup services
• Backup infrastructure costs to protect the TB of data
• Power, cooling and floor space
• SAN, NAS filers, cabling
• Long distance dedicated data circuits
• Costs of array migration (at the end of life for the array)
• Costs of waste
• Lost productivity with long provisioning times
• Etc.

There are new, emerging storage architectures that scale-up and scale-out. These architectures are getting very popular for cloud-enabled services and storage. We also have been looking at Solid State disk drives, and the metrics that help justify the high cost of purchase. Clouds, SSD and scale-up/scale-out architectures drive us to a new metric - $/IOPS/GB.

Now, 3-dimensional metrics are often hard for me to grasp; it seems like I often live in a 2 ½ dimensional world… getting that last little dimension throws me sometimes. Rather than work with a metric with 3-moving parts, I like to lock on the variables, and then use the other 2 metrics to tell the comparative story. A few examples:

1.  People often tell me that SSD or Flash drives are too expensive (and they are to buy) but they ignore the cost to attain a certain IO level. So for this example, say we need a sustained rate of 280 IOPS for a certain application. We can use a serried of traditional disks, with striping across 8 or 12 arrays to produce a certain throughput. The written-to capacity is set by the application, but with magnetic disk media there would likely by a lot of wasted space. In the SSD world, a single disk or perhaps 2 would meet the need. So the measurement would be $/GB (written, not usable or raw) needed to attain the IO rate. You might be surprised that the SSD achieved TCO parity with this metric.
2.  Example 2 is where you need a certain price-per TB point, say $5,000 TCO per TB/year. So with 2 of the variables fixed, the only remaining variable is the performance or IO. You might review and compare 3-4 different solutions that meet this price point, but you may not be happy with the performance, especially at peak times. If you ignore the IO in the $/IO/GB metric, then it will turn out to be the problematic variable in the future.
3.  Finally, you may leave the price to be the variable. You set the total usable or written to capacity. You set the performance with disk type, density, speed, RAID levels, cache etc. Your IO/GB is fixed, but the price and cost will vary with each solution.

Bottom line for me is that 3-moving parts is a little harder to measure than 2 moving parts. Using a 3-dimensional metric is essential to compare and contrast vendors, architectures and total solutions, but I often have to take on 2-variables at a time.