There is no doubt that all-flash arrays are fast. In fact, for many data centers the performance of an entry level all-flash array provides more IOPS performance than they will ever need. But the reality is most data centers are not really constrained by IOPS. Latency and bandwidth are a bigger concern. IT buyers need to re-think why they are buying all-flash arrays.
Quick Definitions: IOPS, Latency and Bandwidth
In our on demand webinar “Understanding IOPS, Bandwidth and Latency” we define these performance parameters in detail, but here are some quick definitions. IOPS, the most overused stat in storage, is how quickly the storage system can deliver or write a piece of information once the system has it. Latency is how long it takes to get that piece of information through the storage system to the point that it can write it, and bandwidth is how much information the system can transfer at a given point in time.
Essentially bandwidth is the first thing we interact, latency is the second and IOPS is the third. For most applications and users, the lower latency of flash (compared to hard disk) is what makes an application “feel” more responsive. Latency is also the one thing that the storage vendor can most directly impact.
Performance Increase Is About Latency Reduction
All-Flash arrays have been on the market for over four years now. In that time each vendor has released systems it claims to be faster than its predecessor. These increases come despite the fact that as flash cell density increases, flash media itself actually becomes slower. What’s the difference? Vendors are improving software so it is more efficient and of course they have built systems that are using more powerful CPUs.
Bandwidth May Be The Next Big Thing
As more organizations tap into the power of the Internet of Things (IoT), Big Data and Analytics, the need to move massive amounts of data to and from storage is increasingly critical. All-Flash helps in two ways. First, all-flash systems have the performance to keep up with high speed networks. Implementing a 10GbE, 16Gbps FC or faster network won’t do any good if the storage system can’t keep up. All-Flash can.
Second, many analytic jobs may consume a lot of capacity but that capacity is made up of billions, if not trillions of small files. The speed at which those little files can be written is severely impacted by a hard disk based system. Here is a situation where all-flash IOPS actually works with bandwidth to make the transfer of these little files, as well as the actual analysis go much faster.
But Density and Power Savings Could be Bigger
For many data centers, we’ve reached the point with all-flash storage where the performance is “good enough”. In other words if the vendor improved IOPS performance (latency may always be an issue) that effort would go to waste. Many data centers simply don’t need more IOPS performance.
There are other reasons however to move to an all-flash system. One of the most interesting is density. Flash is not constrained by the same physics hard drives are. We can fit more capacity in the same space and we can more tightly stack flash modules. Now, we are seeing systems come to market that can deliver PBs of flash storage capacity in a 5U rack space while consuming less wattage than a hair dryer.
All-Flash arrays are about far more than IOPS. Understanding your IOPS, latency and bandwidth requirements, as well as your needs for density, is critical in selecting your next storage system. Join Storage Switzerland and Tegile as we dive deep into these subjects and answer your questions about them.