Download the authoritative guide: Enterprise Data Storage 2018: Optimizing Your Storage Infrastructure
“Go all-flash, young man,” appears to be the current mantra of the storage industry. Vendor after vendor is urging enterprises to ditch hard disk drives (HDDs) in favor of solid state drives (SSDs). You also hear about flash-first strategies that urge organizations to look at flash storage first, last and always.
While this approach makes sense in many cases, what about existing storage assets? And in particular, are there any times when all-flash storage just doesn't make sense?
Here are some possible situations where it might be wise to skip the flash:
Limited available funds are, of course, one reason to avoid flash storage. Sure, flash vendors will say that deploying flash can speed the enterprise and thereby generate income. But all-flash arrays aren’t cheap. If you have budget constraints, flash may not be top of the priority list.https://o1.qnsr.com/log/p.gif?;n=203;c=204650394;s=9477;x=7936;f=201801171506010;u=j;z=TIMESTAMP;a=20392931;e=i
“A little bit of flash in the right place can go a long way, and everybody can benefit from at least a little bit of flash somewhere,” said Greg Schulz, an analyst with StorageIO Group. “Some might say the more, the better. But where you have budget constraints that simply preclude you from having more flash for things such as cold, inactive, or seldom access data, you should explore other options.”
Schulz added, “You can use a little bit of flash to create a hybrid buffer or cache in front of slower storage as a way to stretch your budget. The bottom line on where flash does not make sense is where you are watching your cents, dollars, and euros for apps that simply do not have the need for speed, at least today.”
Some workloads require all the IOPS that can be mustered. All-flash storage may be a no-brainer for demanding and mission-critical applications. But many applications just don’t need all that juice.
"Both all-flash and hybrid systems make sense and serve a purpose for specific workloads,” said Peter Godman, co-founder and CTO, Qumulo. “All-flash systems are cost effective for workloads that are high-IOPS with frequent access to all data, but many workloads have access patterns that focus on only a subset of the data. In this case, having dormant data live on flash is largely a waste of resources.”
Archives and Dormant Data
Most businesses would struggle with spending 3x to 6x the cost to switch from HDD to all-flash storage, particularly for what might only be marginal performance improvements. The economics of all-flash are likely to become more cost effective in the next few years. This will open the door for its introduction into more areas. But there are some areas where all-flash is unlikely to become viable.
“Workloads that have entirely dormant data, like archive are dominated by all-HDD or hybrid systems,” said Godman. “Flash is 5 or more years away from being able to serve these workloads.”
Video and Medical Records
Dell EMC is very much in the forefront of the all-flash storage push. The company's viewpoint is simple: when you don’t have to worry about performance anymore, it opens the door for many new use cases and considerations. It backs this up by explaining how data reduction technology such as compression and deduplication can lower the effective $/GB for all-flash storage. EMC also touts that fact that flash provides better space density, thereby reducing heat, cooling, power and storage management expenditures.
But the company concedes that there are some situations where flash storage doesn’t add up or where entry cost is a barrier, as covered earlier. Another area where flash might not be needed would be archival data, video surveillance or medical records.
“In the case of medical records, data must be kept for a period of seven years,” said Bob Fine, Dell EMC product marketing, midrange storage. “When you multiply that amount of time by the number of users, it can add up to a significant amount of long-term data.”
For these applications, the top criteria is storage density, and the lower cost/GB for long-term storage takes precedent over performance. In this situation, spinning media would be a far better solution, added Fine. Moving data destined for long-term retention to all-flash would be prohibitively expensive.
IDC analyst Eric Burgener concurs.
“If I had to pick the perfect application where you wouldn’t use flash, I might say an archive where you need the cheapest cost for raw capacity for data that is not at all latency- or bandwidth-sensitive, and you need to keep it around for a long time (e.g. deep medical archive),” said Burgener.
Brett Schechter, senior product marketing manager, Tintri, explained that all-flash storage is not necessary when it comes to a strong data protection/disaster recovery (DP/DR) plan and strategy in place. He added that the most important things the storage system must have are snapshots, replication and cloning. Inline deduplication and compression are also essential when it comes to conserving capacity and network bandwidth, but all-flash is not crucial to DP/DR alone.
“In addition, enterprise-grade spinning disks have evolved to be more durable than in the past so that uptime of 99.99999 (seven 9s) is achievable,” said Schechter. “That seven 9s uptime may be more architecturally critical an outcome than raw speed.”
Michel Courtoy, COO of software-defined scale-out NAS vendor Rozo Systems, believes hybrid systems combining flash and HDDs will be around for some time. The transition from HDD to flash is yet another reason why software-defined storage makes sense, he said.
“The price premium for flash is still significant and will stay so for a couple of years so hybrid systems that merge HDDs with flash offer the best of both worlds — high performance and low cost — if combined with the right storage and data management software,” said Courtoy. “Software-defined storage enables users to adjust the flash-to-HDD mix with minimal incremental costs.”
Photo courtesy of Shutterstock.