All flash arrays (AFAs) are really catching on. But with so many companies out there only now joining the party, here are some tips to ease any potential implementation headaches.
1. Factor in Time
ROI calculations are never simple. But those attempting to justify a hefty investment in flash should also factor in time savings as part of the equation. Instead of heading for the more nebulous side such as lower latency causing a certain rise in overall employee productivity – true but hard to monetize with any accuracy – head straight for the storage manager jugular. Do a quick study on storage management or storage administrator time spent on various storage tasks and how many storage hours are saved watching over sluggish disk-based systems or troubleshooting complaints about slow response.
“Performance problems attract (and distract) the most senior storage professionals because performance problems are always urgent and require an in-depth understanding of complex technology interactions across servers, storage, networks, applications and even hypervisors,” said Lee Caswell, vice president of product, solutions and services marketing at NetApp. “Flash eliminates a major performance constraint so that storage admins can spend time on higher value activity. In a time of shrinking IT budgets and staffs, think of justifying flash purchases based on the most recent Gartner data that shows you can reclaim half of your team by deploying flash with a 5-month payback.”
2. Don’t Miss the Point
Jeramiah Dooley, Principal Flash Architect at SolidFire, thinks many are missing the point when it comes to AFAs. He says flash has gotten to the point in terms of performance and cost where it needs to be looked at as more than a point solution for the data center.
“Flash is a strategic change in how storage can be provisioned to support applications so buyers should look for storage architectures that allow them to scale the capacity and performance resources separately, provision them in whatever ratio is required, and protect those workloads at scale no matter how many workloads are being supported,” said Dooley. “The goal isn’t just to use flash and make things faster, it’s to use the consistency of flash in order to change the fundamental model of how storage is used in the data center.”
3. Not Just Throughput and IOPS
Most vendors try to juke it out with raw throughput and IOPS numbers. They throw out benchmarks like confetti. And then another vendor claims they can top those figures. But that discussion may be off point for some flash implementations.
“If you’re running a database, throughput is not your primary concern—it is going to be latency,” said Keith Parker, Violin Memory’s director of product marketing.
Databases, after all, don’t tend to push a lot of IOPS. You can have a huge database, and it’s going to be using 75,000-100,000 IOPS. Where those databases are likely to be challenged, however, is in response times: how long does it take to get a response to a request when you’ve completed a form on the website. Getting that time down is important even though the amount of data is trivial.
“Moving from a disk-based system, you probably get huge latency,” said Parker. “Any move to flash is going to be an improvement, but what kind of improvement are you looking at getting?”
Some AFA implementations may provide 10 milliseconds while others may be able to take that down to 1 millisecond of latency, or even lower. That has a huge impact to users.
4. Application Needs
Anybody who is looking to migrate to all-flash needs to consider if prospective AFAs can support multiple workloads such as their database, web servers, email server, SharePoint and multiple applications on a single array. There are going to be different needs depending on the application mix and overall environment.
“A database is going to need extremely low latency, that’s what it cares about, whereas with a bunch of Web servers its more about throughput and IOPS,” said Parker. “Can the storage meet the needs of each app without having a significant decrease in performance?”
5. Up or Out?
Some AFAs are better at scaling up. Others are more suited to scaling out, and a few are good at both. Scale up is where you start out small with a single AFA and add capacity to that unit, said Parker. Scale out is where you cluster multiple boxes together to increase capacity and performance. Each has advantages and disadvantages related to the environment and workloads. This factor should be given due consideration during product evaluation and implementation.
6. DR Options
Parker also draws attention to disaster recovery (DR). It’s all very well to add an AFA or two, but how are you going to protect that data? Some AFAs make it easy to put storage at two different locations and have the AFAs synchronously mirror to each other.
“That would provide resiliency should any point in the system fail,” said Parker.
7. SSD or CFM?
When it comes to all flash arrays, the battle between solid state disk (SSD) and custom flash module (CFM) is more heated than ever as organizations continue to weigh the merits of each approach into their storage environment. Supporters of SSDs contend that they provide a ready-made form factor that allows them to bypass many flash management challenges associated with hard disk drive, enabling them to concentrate on other aspects of enterprise storage.
Detractors, however, say the SSD approach is hampered by protocols designed for disk. CFM, on the other hand, takes a ground-up approach which integrates the NAND flash into its array along with its own controllers to manage that flash. This typically results in better performance, said Parker.
8. Consider Other Systems
Dan Leary, vice president of products and alliances at Nimble Storage, cautions users that an AFA is not necessarily the be-all and end-all of performance. While flash undeniably provides an incredible boost to performance, Leary said 54 percent of all potential application slowdowns — the kind most often attributed to storage — are not caused by the storage system.
“When considering an all flash array, understand that performance isn’t a cure all,” he said. “There are other factors that affect application performance such as hosts, networking, interoperability and failure to follow best practices when addressing issues.”
Photo courtesy of Shutterstock.