![Industry Perspectives Industry Perspectives](https://eu-images.contentstack.com/v3/assets/blt8eb3cdfc1fce5194/blt79a1824cfad941c4/664ba129be90a086233eb8e0/Industry_Perspectives_DCK_light.png?width=700&auto=webp&quality=80&disable=upscale)
Insight and analysis on the data center space from industry thought leaders.
Storage Wars: Dispelling the Myth of Flash Economics
There are new metrics for measuring flash costs and taking a look at them can illustrate that the return on investment (ROI) that can be achieved with this new technology, writes Walter Hinton of Virident.
July 23, 2013
Walter Hinton is the Sr. Director of Field Marketing at Virident, a performance leader in flash-based storage-class memory (SCM) solutions.
Walter-Hinton_Virident_tnjp
WALTER HINTONVirident
While flash storage is generating a lot of positive buzz for its improved response times and low latencies, it suffers from a reputation of being very expensive. This is true, if the cost of flash is measured like that of traditional spinning disks (Cost/MB, Cost/GB, Cost/TB). However, measuring flash as a capacity commodity is not practical, as flash delivers on very different value propositions (IO and throughput) that can only be achieved by using RAID and lots of hard drives bundled together. Let’s take a look at new metrics for measuring flash costs and illustrate the ROI that can be achieved with this new technology.
Storage Causes the Bottleneck
With multi-core processors and IO-hungry applications in today’s enterprise and web-scale organizations, the performance bottleneck has moved from the server to storage. Given that a typical hard drive can deliver 175 to 210 IOPs and a PCIe card can sustain more than 325,000 IOPs (using the same measurement technique), one could argue that it would take 1,547 SAS drives to achieve the same performance that can be delivered by one card. Given that each of the SAS drives could have as much as 600GB of raw capacity, you could easily end up buying 928TB of capacity to hold a database that actually needs less than 2TB to host the application, log files and meta data.
From a disk drive perspective the cost/GB for that device is less, approximately $0.10/GB, while a PCIe card could be as much as $8/GB, but if you’re seeking performance, paying $16,000 for a 2TB flash card vs. $92,800 for SAS drives is a significant difference.
A New Measurement
What about considering a new metric for flash technology? Since the goal of flash is to deliver application acceleration through low latencies and high IOPs, perhaps we should consider the cost/IOP as a more relevant metric. If a 2.2TB card can deliver 325,000 IOPs at $16,000, there’s a cost/IOP of $.05/IOP. Considering the above example of 1,547 drives at $92,800 to achieve the same level of IOPs, the tables are suddenly turned and the cost of traditional storage looks like $0.28/IOP, more than a five times difference in cost.
Virident
Another big driver of cost in the enterprise and web-scale environments is power. As one might imagine, flash really shows off when we look at cost/watt. With 1,547 drives each consuming anywhere from 7 to 17 watts per hour each (varies by supplier), a scale-out environment can easily exceed 10,000 watts per hour. This compares to a PCIe card with equivalent IOPs consuming 25 watts per hour at maximum use. Given an industry average of $6/watt/year, that’s a power consumption differential of $60,000/year for the hard drive solution compared to $150/year for a card.
More and Faster Drives
Of course, all of the above does not take into consideration storage arrays that have massive amounts of RAM and hundreds of drives to meet the IOPs challenge. But these systems cost hundreds of thousands of dollars. Take, for example, an enterprise array targeted at high-end applications like enterprise database and web-scale organizations. This array consumes more than 400 watts/hour (without fully populated drives) and boasts 39,041 IOPs with built-in SSDs and SAS drives. It costs more than $115,000 when populated with just 17.8TB of capacity. The key value proposition of the array is the ability to add SSDs, fast SAS drives and capacity drives for a complete storage hierarchy. But the fact remains that if IOPs is a priority, you need more fast drives, so costs go up and power consumption increases proportionally.
We must dispel a myth about storage costs, particularly for flash. As storage professionals, we have been taught that the cost of storage drops by 40 percent per year. This is not the case with flash (at least not today). Costs will come down as more technology is deployed and as the market grows, but for the past five years, we have seen only a modest decline in flash costs (5-10 percent per year). Don’t expect flash to be cheaper on a pure dollars/GB basis than HDDs any time soon. However, as we’ve discussed, flash shouldn’t be measured in dollars/GB but in dollars/IOPs, and that is where the cost savings are seen.
Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission process for information on participating. View previously published Industry Perspectives in our Knowledge Library.
About the Author
You May Also Like