SSD Can Now Achieve HDD Price Parity

This article, SSD Can Achieve HDD Price Parity…NOW, has been written by George Crump, senior analyst at Storage Switzerland, LLC. Note that Skyera Inc. is a client of Storage Switzerland.

      
George Crump

"When will the data center’s storage infrastructure become predominantly solid state based?" is a common question regularly posed to analyst firms like Storage Switzerland. While there are many factors, the most realistic answer is "when flash systems achieve raw price parity with HDD based enterprise systems". At that point there is no longer a cost justification that needs to be made and the all solid state data center begins to emerge.

Today when flash vendors compare pricing to HDD there is a lot of ‘fuzzy math’ applied. For the ‘all solid state’ data center to emerge NAND flash systems will need to achieve true price parity based on $/GB, not $/IOP. The $/IOP calculation, which has become a standard method for flash system vendors to cost justify their wares, is problematic. It requires that additional justification be made to explain the significant $/GB delta between flash systems and enterprise disk systems.

Clearly there are already applications and environments that can make this leap, since flash systems are being purchased today. But the effort required to make the justification leaves many applications on the outside looking in the solid state performance window. These are applications that could see benefit from the performance solid state offers but can’t quite justify it.

There is also the reality that no matter how solid the $/IOP math turns out data center managers simply can’t justify the additional expenditure in today’s budget environment. This is similar to a new car buyer wanting to purchase a hybrid to save money on gas but can’t afford the upfront investment, even though it would be more economical in the long term.

How Are Flash Systems Justified?
The increased cost of flash systems is typically justified by comparing their cost to disk systems configured to deliver the same levels of performance. The most ideal situation is one where the data center is facing an investment in hundreds of HDDs in an enterprise storage system. Those drives are then typically short stroked (partially formatted) and widely striped (high drive count RAID groups) to achieve the maximum performance that a demanding application or environment needs. This approach leads to under provisioned hard drives that waste capacity and consume more power than they should.

In many cases a storage system like this does not need the capacity that is provided by the hundreds of HDDs. By comparison the flash system now looks affordable. In a significantly smaller package it can provide the same or better performance, deliver the same capacity and provide a significant savings in space, power and cooling. But again, there are a finite number of applications where such high drive counts would be required and where flash systems would have the price advantage.

Price Parity Methods
Flash system vendors will also leverage a series of technologies to achieve an artificial price parity or to at least blunt the impact of SSD pricing. The first of these is the very common practice of caching or tiering. In these cases a much smaller SSD area is set up to hold only the most active data. Essentially this method justifies the SSD investment by auto-qualifying the data center’s data and moving it to the available SSD automatically, while keeping the rest of the data on HDD.

The caching and tiering methods work fine until there is a cache or tier miss, a situation where the wrong data is in the cache/tier and the right data is on the HDD. When this happens a sophisticated data exchange needs to occur between the wrong data on the SSD and the right data on the HDD. The performance impact of waiting for this exchange can negate the advantages of the SSD investment altogether, especially if the exchange occurs too often. Frequent exchanges can be caused by widely active data, poorly designed data analysis software or an under-sized SSD storage area.

All-flash systems present don’t suffer from the cache/tier miss problem, but do have a cost justification problem. To work around this many flash systems are applying storage efficiency technologies to SSDs. This is done through the use of deduplication and compression, which eliminate redundancy between and within data segments. These technologies, popularized in disk backup storage systems, allowed disk technology to compare more favorably to tape based technology. While they do not achieve the same efficiency rates in primary storage as they did in backups (typically 5:1 vs 20:1), these technologies have made their way into mainstream data center storage systems. The use of storage efficiency is especially attractive in flash systems, where because of the price premium, any additional savings pays larger dividends.

Justification Still Required
While these methods have made the flash system’s justification easier and brought solid state within the reach of more applications in the data center, they still require that the justification be done. Also, each of these techniques requires the right conditions be in place in order to make that justification work. This means either the ideal comparison to hundreds of hard drives, a data center that has a small, consistent, cache-friendly data set or one with highly redundant information, ideal for deduplication/compression.

Essentially these techniques make the $/GB disparity less of an issue so that $/IOP can be more attractive. However, the capability now exists to make the $/GB a non-issue, even without the above conditions being present, but it may require changes in the way flash systems are designed.

How to Achieve $/GB Parity
The first step in flash systems achieving 100% price parity with enterprise, HDD-based storage is for flash system vendors to completely embrace vertical integration. Many of these vendors, in order to speed delivery to market, have leveraged solid state flash memory designed to look like traditional HDDs and placed those drives into standard off the shelf storage servers. These manufacturers are counting on software that can provide deduplication and compression to address the price disparity.

While the use of flash in drive form-factors does speed up time to market and reduce initial inventory for the system vendor it adds to the cost of the system. Clearly the manufacturers that provide the flash drives to the system vendors need to make a profit on the sale. That profit adds to the cost of the system, a situation called ‘margin stacking’. In addition, flash as a drive does not take advantage of memory based storage’s capabilities for density and power efficiency. As a result these systems can draw as much power and take up as much space (on a shelf by shelf basis) as a hard drive system.

If the vendor is using solid state in a drive form factor then they also need to leverage that vendor’s controller to handle functions like flash wear management and reliability. Even vendors that chose to not use flash as a drive solution often look to another vendor to provide the controller function. As before this leads to margin stacking and lack of innovation.

We are seeing flash system vendors like Skyera become more vertically integrated. Ideally these vendors will begin to just buy the raw NAND flash from manufacturers like Toshiba, Samsung and Micron. Then they will develop their own flash layout strategies and not be tied to a legacy drive configuration. Finally, they will develop their own flash controller. Doing so virtually eliminates margin stacking and leads to denser configurations, another key component of price parity. Fewer shelves and power supplies lower the overall cost of the system and reduce data center floor space and power consumption in the process.

When a flash system vendor develops their own controller technology it also gives them freedom to advance their controller functionality so it can provide enterprise class reliability to low cost consumer grade MLC flash. But the innovation required for flash to HDD price parity has to come from innovation up and down the technology stack, including ECC algorithm optimization, new and innovative RAID strategies.

In addition, these flash system vendors can now more quickly embrace shrinking lithographies that offer better flash price points. Skyera for example can leverage consumer grade MLC and 19nm to 20nm technology to access flash storage at $3/GB raw. Add deduplication and compression, which Skyera has, and the price per GB moves closer to $1/GB.

Conclusion
If data centers had the choice they would choose solid state for 100% of their primary storage needs. The technology is power- and space-efficient and of course, brings unprecedented performance to today’s demanding applications and workloads. To make that choice flash systems will need to achieve true price parity with HDDs. This price parity will have to come from tighter integration and innovation up and down the technology stack. The more of the stack that the flash system vendor owns the faster they will be able to bring price parity to the data center.