BACKGROUND IMAGE: iSTOCK/GETTY IMAGES
All-flash arrays have become more common over the past two years, but the majority of flash storage today is still on hybrid arrays that place data on flash for applications that need high performance, while other data remains on traditional disk. The debate about which approach to choose is ongoing. SearchSolidStateStorage.com managing editor Ed Hannan and independent storage expert Andrew Reichman recently discussed the topic.
Which is better: hybrid arrays or all-flash arrays? Why?
Andrew Reichman: As with most questions, it depends. All-flash has the benefit of not needing to determine which data is hot in order to support tiering. Without tiering, all data is fast, so you don't get any variability of latency for the application. With tiering, if the system or admin gets it wrong, you would have a lag in performance that could be noticeable to users and to apps.
You also have the potential for using system performance resources to accomplish the tiering, which can stretch the capabilities of disk-based systems. However, flash remains much more expensive than disk on a [cost pre gigabyte] basis, so using disk and flash together can be a much more economical scenario. Many modern disk systems have auto-tiering tools that allow the system to determine which data should live on flash and which data makes more sense to keep on a disk tier.
The effectiveness of auto-tiering will likely make or break a hybrid array. If it's effective, you might see a small amount of flash in a system providing a huge boost in performance, with 75% or more of the I/O being served by that small flash tier. I have talked to organizations that are experiencing this result.
When does it make sense to use a hybrid array? When is it better to use an all-flash array?
Reichman: A hybrid array makes sense for general-purpose applications that have lots of cold data, some hot data, and just need a boost over what they get from a disk-based system. An all-flash array might make sense for a specific application that depends heavily on high performance, for environments that can share and tier their data based on some kind of internal tiering -- strict archiving policies, clear working sets, etc.
Some all-flash systems make heavy use of deduplication to shrink the amount of capacity necessary and make up for the cost difference. In this case, apps with highly redundant data might be good candidates for all-flash. If it's extremely performance-sensitive, then it might need all flash. If it's a general-purpose app, with large portions of data that don't get accessed frequently, then all-flash may well be overkill.
Do hybrid arrays deliver better bang for the buck?
Reichman: Probably. For cost-sensitive environments that need moderate performance improvement over large capacities, hybrid is probably a better fit.
Who are some of the leading array manufacturers and what can you tell us about their products in terms of price, features, functionality, etc.?
Reichman: Most every disk vendor today has hybrid -- EMC, NetApp, HP, HDS, IBM, Dell, Nimble, etc. Of those, EMC and Dell/Compellent are probably the leaders when it comes to auto-tiering, which makes a big difference in the effectiveness of a hybrid system.
For all-flash players, Pure Storage, Nimbus, SolidFire, EMC XtremIO, Cisco/Whiptail, Violin Memory, Skyera and Kaminario are the main players, with NetApp, Dell, HP and IBM also touting all-flash products. None of these have huge revenue, and each has varying benefits and drawbacks. Most all-flash systems are fairly small, requiring administrators to figure out which application data sets to put on them, deciding which apps have access to all-flash, while the rest of the data is stored on a separate disk system.
Where do you see this debate headed? Will one emerge as the standard or is there room for both hybrid and all-flash arrays?
Reichman: With more virtualization and more software-defined storage, I think the world will gravitate towards hybrid systems with more flash capacity, allowing entire application data sets to reside in flash, but with lots of cheap and deep capacity on hand to store colder, infrequently accessed data -- of which there is a tremendous amount. Certain verticals may stick with all-flash as part of their workflow, but it just doesn't seem likely to overtake disk usage any time soon.
Hybrid arrays focus of enterprise flash usage
Comparing hybrid arrays, all-flash arrays
Is consolidation coming to all-flash array market?