All-flash storage arrays are becoming well known, and their adoption rate in IT shops is increasing. At Demartek, we are even hearing that some large IT shops are including all-flash array storage in their future purchase plans, and that all-flash arrays are becoming their standard for Tier-1 storage platforms for mission-critical active data.
This article is the first in a series that walks you through the buying process for all-flash solid-state storage arrays. You'll learn why IT shops are considering all-flash arrays for Tier-1 storage and which applications benefit the most from flash performance improvements.
The progression of all-flash storage arrays
Shops are now considering flash storage arrays for Tier-1 use because designers have added management and other features that put them on par with today's external hard disk drive (HDD) storage systems. Initially, all-flash arrays lacked enterprise-class capacity and functionality; they were used to accelerate performance of niche applications. Today's all-flash array capacities, physical characteristics, application, features and endurance now rival those of market-leading HDD arrays.
Individual solid-state drives (SSDs) used in some enterprise all-flash arrays are available today in 1.6 TB or 1.9 TB capacities. These exceed the capacities of enterprise 10,000 rpm or 15,000 rpm HDDs. Although today's 7,200 rpm HDDs are available in larger capacities, SSDs are gaining capacity fairly rapidly.
The physical characteristics of all-flash array storage are also becoming appealing for IT managers. Many all-flash arrays consume significantly less than 1,000 watts per 2U storage system. In many cases, data centers are finding that increased amounts of power are simply not available from the local electric utility, so any technology that reduces power consumption is beneficial.
Because all-flash arrays consume less power and do not require as much cooling as HDDs, they produce less heat overall, thereby reducing the air conditioning requirement in a data center. Also, many all-flash arrays run quieter than HDD arrays in the same amount of rack space.
There is an interesting workload trend emerging with all-flash array storage. Initially, IT shops may deploy an all-flash array for a single workload or application. They often notice an all-flash array performs very well for a single application and that there is room for growth of that application in terms of performance. As a result, these shops begin to add a second workload to the same all-flash array, then a third workload and so on.
For example, we have run multiple online transaction processing (OLTP) and data warehousing workloads on the same all-flash array and obtained very good performance. We have not been able to run those same multiple OLTP and data warehousing workloads on a HDD array of the same capacity and achieve the same performance.
Many of today's all-flash arrays have incorporated advanced features such as compression, data deduplication, thin provisioning, replication, snapshots and encryption technologies. Some of the data reduction technologies, such as compression and data deduplication, help to drive down the price by increasing the effective usable capacity for a given amount of raw flash.
However, this also can become a point of confusion because the various vendors may not compute their prices on the same effective data rate, and the effective capacity varies by the type of data stored. There is also some debate among product vendors about whether one should perform compression before data deduplication or vice versa. The optimal answer depends in part on the architecture of the particular all-flash array.
In general, it seems that management of all-flash solid-state storage arrays is simpler than traditional HDD arrays. In older HDD arrays, there were limitations on the way logical volumes could be created. Disk groups had to be created with a fixed number of disks in the group, and a specific RAID type associated with that disk group. Storage administrators had to keep track of these disk groups, and in a large array, this could be time-consuming. It was also a serious amount of work to change the disk group. Most all-flash arrays today use a variation of wide-striping or variable-striping that allows volumes to be built across many or all of the drives or flash modules in the system.
The endurance of all-flash arrays has been a perennial topic of discussion. With improvements in wear-leveling, error correction code and other related features at the flash controller level, many of these endurance-related issues have been solved. Products have been in the field long enough to see that failure rates are quite low; in some cases, lower than HDD failure rates. This is why it is not uncommon to see five-year warranties from all-flash array vendors.
Pricing discussions are always interesting when it comes to all-flash arrays. A couple of years ago, $5 per gigabyte (GB) seemed to be a target price for all-flash array storage. This price has dropped due to capacity improvements in NAND flash technology and advances in data reduction features such as compression and data deduplication. We are now hearing of prices dropping to approximately $2 per GB for effective usable capacity, assuming a fairly large capacity model is purchased. I have even heard that prices are expected to get down to $1 per GB within the next year or two for the larger-capacity models of all-flash arrays.
One of the selling points is for customers to buy enough flash for today, then purchase capacity upgrades in a year or two as the prices drop. This can be done by planning to purchase a certain size of drive or flash module now, and then buy a larger size of drive or module next year.
As an independent test lab, we spend the majority of our time measuring various aspects of performance for servers, networking and storage systems. For storage systems, we usually capture three basic metrics of storage performance:
- Input/output operations per second (IOPS)
- Throughput measured in megabytes per second (MBps)
- Latency measured in milliseconds (ms) or microseconds (µs) in addition to other metrics from the application host server.
When we tested all-flash arrays, the first thing we noticed compared to HDD-based arrays is the significant difference in the three basic metrics and performance consistency. Although it is workload dependent, all-flash arrays generally have more consistent overall performance than HDD-based arrays, and we especially see this in the latency measurements.
Many workloads -- OLTP database, virtual desktop infrastructure and Web server workloads, to name a few -- benefit from latency improvements. When these workloads are moved to all-flash arrays, the end-user experience is dramatically improved because response time is reduced, and the increased performance is consistent. Many all-flash arrays can reduce average latency to less than 1 ms, depending on the workload.
Most transaction-based workloads are also sensitive to improvements in IOPS. Individual HDDs can deliver hundreds of IOPS, while individual SSDs can deliver thousands of IOPS in the same form factor. When these types of drives (2.5-inch) are placed in an all-flash array, it is not uncommon to obtain 100,000 to 400,000 IOPS in a 2U system, depending on the workload.
Other all-flash arrays use flash in different form factors, such as PCI Express (PCIe) cards or some proprietary form factors. These are often called modules. Many of these non-drive form factors have a higher performance per module than SSDs because they use interfaces such as PCIe. Several of these systems deliver even higher IOPS than the drive form-factor systems. For some customers, this is more than enough to handle their transactional workloads. For other customers, this type of performance opens up new application opportunities.
Some large customers are taking advantage of all-flash array storage because of the increased throughput available. Some of the extract, transform and load workloads, analytics processes, application-specific replication tasks, backup tasks and streaming tasks for large databases can be performed in significantly less time. We typically see time savings measured in hours or sometimes days with all-flash arrays, depending on the specific workload. When repeated daily or weekly, these time savings can help to offset the higher cost of a storage system.
All-flash arrays have been in the field for a while, and with many of them now offering the advanced features and high performance required by enterprises, it's hard to find a reason not to purchase one. All-flash arrays are also much more affordable than they were a few years ago, and their reliability is quite good.
About the author:
Dennis Martin has been working in the IT industry since 1980. He is the founder and president of Demartek, a computer industry analyst organization and testing lab.
What is flash storage and when to deploy it
Flash array storage rising in popularity
How to interpret vendor's all-flash array storage performance claims
Technologists discuss the future of enterprise flash storage
Flash array storage factors to consider before you buy