There has been ongoing industry speculation about whether solid state storage will become the predominant medium in enterprise storage systems. The assertion is that the flash SSD advantages over HDD of higher IOPS, reduced power, reduced cooling and lower failure rates are so compelling flash SSD will eventually replace the HDD within storage systems. But analyst conventional wisdom has been (and continues to be) that flash SSDs cost...
way too much per usable terabyte of capacity. Historically, that has been true.
This led storage system vendors to develop creative workarounds emphasizing flash SSD advantages while minimizing their costs. The workarounds either utilize flash SSDs as cache or as tier 0 storage with automated tiering software. These workarounds are often referred to as hybrid storage.
Caching accelerates reads and, in some cases, pins important data within the flash SSD cache. Flash caching is primarily based on intelligent analysis of read frequency. All writes are passed through directly to HDDs. Based on frequency of reads, data is placed into the SSD cache. When read frequency decreases, the data is removed from cache. Issues with flash cache intelligent caching include policy limitations and amount of cache. Once again, too little cache leads to increased cache misses and redirects to HDDs, defeating the purpose of the flash cache.
Tier 0 storage utilizes flash SSDs as the primary storage target for applications requiring higher performance -- specifically, higher IOPS. Effective use requires some form of policy-based automated storage tiering (i.e., no admin intervention is required to move data from one storage tier to another). Automated tiering is based on data value. As data ages, its value decreases and is less frequently accessed. The issues with Tier 0 automated storage tiering are the cost of automated storage tiering software and system complexity. Software costs can greatly offset any savings from the reduced numbers of flash SSDs. Automated tiering is appealing to storage admins; however, implementing requires skill and attention to detail. Details such as making sure both snapshot and data protection policies match the data regardless of the tier they've been moved to.
An underlying reason for limiting flash SSDs in storage systems is performance limitations in storage controllers and back ends. These rapidly become performance bottlenecks when too many flash SSDs are implemented. They were not architected with flash SSD performance in mind. As a result, hybrid-optimized flash SSD/HDD storage systems have become the new norm.
But what if the upfront cost of usable flash-SSD capacity was equivalent to SAS 10,000 rpm HDDs? And what if the storage system was able to fully utilize the performance of large numbers of SSDs? The answers to those questions are emerging from storage system startups like Astute Networks, Nimbus Data, PureStorage, Skyera, SolidFire, Tegile, Texas Memory Systems (now part of IBM), Violin, Whiptail, XtremIO (now part of EMC) and others.
Each of these vendors offers (or, in the case of EMC-XtremIO, will offer in the near future) a 100% flash SSD storage system. They each have architected their storage controllers to take full advantage of flash SSD performance. Some utilize a scale-up design, such as Nimbus and PureStorage. Others use scale-out, including Skyera, SolidFire and EMC-XtremIO. And some are more specifically market-focused, such as Astute with VM-optimized storage; SolidFire with guaranteed performance SLA management for cloud service providers; TMS and Violin with very high performance for the HPC market; and Whiptail for the low end of the mid-tier market.
Upfront costs per gigabyte (GB) typically range from a low of $3 to a high of $15 for all-flash SSDs. This is before the storage systems apply inline deduplication and compression. All-flash SSD system performance allows inline dedupe and compression to be applied with nominal application performance impact. (This is also true for hybrid storage systems and not true for HDD storage systems.) The result is a cost per GB ranging from approximately $1 to $5.
Compared with hybrid flash SSD/HDD systems, upfront costs per GB commonly range from $1.6 to $24 per GB before inline dedupe/compression, and approximately $.54 to $8 afterwards. Traditional HDD system upfront costs per GB regularly range from $.5 to $15 depending on the mix of SATA, NL-SAS, SAS, or FC HDDs.
It's when the power and cooling costs are factored in that the 100% flash SSD systems costs become very competitive. Flash SSD storage system power and cooling consumption ranges from approximately 50% to 80% less than HDD systems. Power and cooling costs per year for HDD storage systems in recent history have been tracked at roughly the same as the upfront hardware costs. A comparison of the power and cooling savings of 100% flash storage systems and hybrid systems depends on the mix of SSDs and HDDs and whether or not the HDDs have a "spin-down" capability. For the sake of argument, we will generally assume that the power and cooling savings is 25% less than it is versus 100% HDD systems. So if we assume the average SSD system's power and cooling savings is generally around 67% or two-thirds versus a HDD system, and 50% less than a hybrid system (75% x 67% = 50%), and that upfront cost per SSD GB is approximately $5 (based on product actually shipping), general TCO comparisons get quite interesting.
The average SSD system three-year TCO per usable GB gets very competitive as upfront hardware costs drops to $5 or less. Granted, there are a lot of assumptions in this comparison, but it shows that the SSD-HDD pricing gap has closed faster than most have anticipated.
Which brings us full circle back to the original question: With the impediments to 100% flash SSD storage systems rapidly disappearing, has the paradigm shift begun? Based on recently reported vendor sales, the answer would appear to be yes. One of the more mature 100% SSD storage system vendors, Nimbus Data, has reported 500% year-over-year growth. Others -- such as Astute, Violin and Whiptail -- are reporting sales growth well in excess of standard storage systems. These reports must be taken somewhat with a grain of salt. All of the vendors are private and don't have to provide any details. And all of that growth is coming from a very small base. They are, however, data points that point in the right direction. It will take some time to see if this is the start of the predicted paradigm shift, or an anomaly.
In the end, 100% flash storage systems, with their simplicity, competitive TCO and much higher performance, gives these systems credibility for any new storage acquisition. One more note: Many traditional storage vendors will charge considerably more for 100% flash storage systems than the emerging startups. Don't be limited only by the traditional storage vendors.
About the author:
Marc Staimer is the founder and senior analyst with Dragon Slayer Consulting in Beaverton, Ore. Marc can be reached at firstname.lastname@example.org.
Dig deeper on Solid state cache appliance implementations