All-Flash arrays are storage systems that are 100% solid state and give a complete set of robust features like snapshots and replication, similar to traditional hard-drive based storage systems. While their performance enhancing potential is very appealing, All-Flash arrays have two basic challenges to overcome. First is the cost; an array populated entirely with (flash) storage capacity is about 10X the price of a comparable hard disk device. The second challenge is what to do with data that really shouldn’t be on an All-Flash array because it can’t take advantage of the performance that an All-Flash system can deliver.
Overcoming The Price Challenge
Overcoming the price challenge is the primary focus of All-Flash Array vendors and covered in detail in the article “Overcoming The All Flash Storage Challenge…Cost”. To address cost All-Flash arrays typically include features like deduplication and compression to maximize capacity use. Other manufacturers have focused on high levels of vertical integration to drive out the cost associated with using and paying multiple vendors for the components that end up in their arrays. As a result of these efforts we’ve seen several All-Flash array vendors now claim price parity with enterprise disk based arrays.
Most of these pricing claims have at least two asterisks by them. The first asterisk is that they make this claim based on effective capacity, after deduplication and compression have done their work, not raw capacity. An exception is the vertical integrators who tend to base their claims on native (uncompressed) capacity. The challenge with including compression/dedupe in the price comparison is that every data center’s return on the dedupe/compression investment will vary.
The second asterisk of the price parity claim is that the hard drive system that most are comparing to is an enterprise class, high performance system, in other words a storage system with 15K RPM hard drives from a top-tier vendor. There is nothing wrong with that comparison as long as you have such a system in your data center and you intend on replacing it with an All-Flash array.
Even considering these two asterisks our findings show that the claims of price parity are essentially fair, especially when you consider that you should be able to do more with an All-Flash storage system than one based on hard disk drives (HDDs). “More” means having the ability to support more virtual desktops, more virtual servers or more simultaneous database users.
The challenge that All-Flash storage systems face though is that most hard drive based systems are not exclusively loaded with 15k RPM drives. Most will have a mixture of 15K, 10K and even high-capacity 7200RPM drives. Increasingly, legacy systems will also have a tier of SSD. The point is that legacy storage systems will give options for where administrators can place data depending on its performance demands. All-Flash arrays do not. The data placement challenge is the second challenge (after price) that needs to be overcome.
Overcoming The Data Placement Challenge
With the inability to place data on something other than Flash, the All-Flash arrays is like a hard disk array that only accepts 15K RPM disks. There is no secondary performance options in the array that allows the user to save money. If this can’t be addressed the All-Flash array is going to be bought to solve niche problems, such as high transaction databases and VDI environments. The All-Flash vendors (the ones that have customers) tend to have most of their customers in these categories.
There are enough data centers that have this need that the All-Flash array vendors care doing OK by supplying that demand. Most of All-Flash array vendors though have their sights set on being used more broadly in the data center. To do that we think they will need the ability to ease data movement to and from their platforms or the ability to add hard disk to their systems.
Facilitate Data Movement
If All-Flash systems can integrate with or develop their own data movement capabilities this would increase their attraction to many more data centers. Instead of replacing the existing hard disk systems they could compliment them and even extend the life of them. The technology to live-migrate volumes of data between storage systems is available from VMware with Storage vMotion and from at least two storage hardware vendors which have that ability as part of the storage system itself. Integrating this at the storage system level is ideal, but in all three examples you are locked into the vendor’s solution. It must be a VMware virtual machine or each storage system being migrated to and from must be from the same vendor.
Beyond the vendor-agnostic migration ability, the All-Flash vendor would need the ability to run analytics on the storage systems’ volumes so that decisions could be made about what should or shouldn’t be on the All-Flash array. This is not a cache but a static placement of data based on the analytics facilitated by live, transparent migration of volumes when needed. Probably the biggest challenge would be how to tell the attached server of the new location of the volume. Each of these would be major developmental undertakings especially if the vendor tried to do it in a way that would compliment any existing hard disk storage.
In the article “What is Hybrid Storage” Storage Switzerland describes and explains the Hybrid Storage market. The general assumption is that the hard disk is the dominant storage type with the goal being to use as little flash as possible. The risk, as pointed out in that article, is what will performance look like when there is a cache miss, meaning that the data can’t be served from cache?
What if Hybrid Storage is turned ‘on its head’? Instead of a disk-based storage system that adds some flash capacity, what if it was an All-Flash array that added some hard disk capacity? This could even start out as an All-Flash storage system with the ability to add hard disk at a later date. Similar to All-Flash arrays most Hybrid Storage systems have deduplication and compression (although few do it across both Flash and HDD tiers), and they have the full storage feature set that we have come to expect from an enterprise array.
These systems are all based on a caching implementation to move data and take advantage of SSD performance. They would need to work around cache’s dependency on a larger, slower tier while/if they were in an all-flash state. Reality is that they typically would not be used that way, most would have a hard drive tier. What we are really suggesting with Hybrid All-Flash is a flash area that is much higher as a percentage than previously considered. Instead of sizing the SSD tier to be 10% of data, maybe size it to 25%. Doing so would virtually eliminate the possibilities of a cache miss while still keeping costs under control.
There are some areas where some Hybrid-Arrays would need to improve if they are going to compete against All-Flash Systems. A good example is dealing with write I/O. Hybrid systems would need the ability to send all inbound writes to the SSD flash tier; many do not. Another is developing a way to pre-warm the cache when older data on the hard disk tier suddenly becomes active. Finally, they will need to be efficient in their resource requirements so that data movement between the two tiers does not impact performance.
In an upcoming article “Hybrid Storage vs. All-Flash Arrays” Storage Switzerland will look at the difference in these two platforms to help you decide which makes the most sense in your data center.