Flash data storage has become the darling of enterprise technology, largely because of its speed. Flash storage systems operate at 400 times the speed of conventional spinning magnetic disks, which makes them ideal for applications such as virtualization and big data analytics. Of course, flash storage is still more expensive than magnetic media, and it’s certainly not a solution for all enterprise data storage needs, but it can be incredibly valuable if you understand where and when to use flash storage arrays to their best advantage.In the realm of enterprise data storage, flash arrays are far outselling other types of data storage systems. Solid-state drives (SSDs) are outselling backup and archiving drive systems two to one and are far ahead in sales compared to hyper-converged infrastructure, software-defined storage, and big data storage. SSDs are becoming increasingly popular in servers for applications that are IOPS-intensive, and as flash technology continues to drop in price, it is becoming more attractive for enterprise data storage.
What’s driving demand for flash data storage is the need for speed. New computing paradigms such as virtualization and cloud computing are creating demand for faster data storage systems, but there’s more to flash than keeping up with the I/O demand.
Understanding the Flash Difference
All-flash storage arrays use solely solid-state storage—typically NAND flash with non-volatile memory. Flash is commonly used to either replace or supplement hard disk drives (HDDs). Flash is also preferred because of its read/write speed.
In an all-flash storage system, the controller has been optimized to accommodate faster data speeds. Where HDDs can deliver 200 to 400 IOPS, SSDs can deliver 8,000 IOPS. As latency drops, the controller becomes the bottleneck, so by using a distributed controller architecture, you can accommodate more data streams and handle data storage services using separate processors.
Flash controllers also have to deal with garbage collection. SSDs must be overwritten as blocks rather than bits, so entire blocks have to be erased prior to new data writes. That means the controller has to manage additional data copy steps.
Due to the cost per gigabyte for flash storage arrays, enterprise architects think twice before adopting SSDs for general enterprise use. However, that perspective is changing. When you factor in the cost of data reduction technologies and deduplication, all-flash data storage can make more economic sense because the high IOPS performance makes SSDs more efficient than HDDs. And because the cost per gigabyte of flash is higher, every byte reclaimed through data reduction provides a higher return per byte than with HDDs.
When calculating the cost and returns of flash storage arrays, it all depends on the enterprise application.
Considerations for Enterprise Flash Storage
While read/write speed may be the primary criterion for flash storage arrays, it’s not the only one. Flash data storage should be used where appropriate in order to simplify data storage and infrastructure design. When they’re not used for the right application, flash arrays can actually make system design more complex.
There are a number of factors to consider beyond IOPS when shopping for the right storage solution:
First, there is storage management. Don’t rely solely on vendor claims of lightning-fast IOPS performance. Make sure that the array also comes with the management features that you have come to expect, such as snapshots, encryption, data replication, and quality of service. Speed won’t be as valuable if you can’t effectively administer data storage.
Assess whether the flash array can deliver continuous availability. If you are investing in a flash array, then the data stored have to be valuable, which means they have to be accessible. Downtime is unacceptable, so look for features such as active/dual active controllers, automatic failover, multipath I/O cabling, and RAID levels that can handle multiple SSDs failing without data loss.
Does the flash array deliver consistently low latency? When measuring data storage performance, the terms “response time” and “latency” tend to be used interchangeably, which can be misleading. It’s one thing to talk about milliseconds or microseconds of response time under a light load. It’s another thing to measure latency as the enterprise throughput increases. Measure latency across multiple loads and scenarios in order to see if you get consistently low latency.
Do You Really Need Flash?
Be sure to ask yourself if a flash storage array is even necessary for the target infrastructure. Conventional HDDs can produce tens of thousands of IOPS, which is fast enough for most enterprise applications. Combining HDD and SSD in a hybrid array can give you the higher performance of flash for hot data while using less expensive disk storage for data that are used less frequently.
If there are a higher number of random read/writes, as with virtualization and transactional databases, then an all-flash array makes sense. Traditional enterprise services such as email, collaboration, and Web services don’t get the same benefit from the speed of flash arrays and are better served with flash cache or supplemental SSD storage.
However, there is no doubt that flash storage arrays are becoming more accepted and more affordable for mission-critical enterprise applications. Advances in compression and data deduplication mean that users can purchase less flash storage capacity and still accommodate more data in a high-speed array. The overall capital expenditure for flash storage arrays is dropping, making them an attractive data-center investment, especially for enterprise customers with expanding virtualization and cloud-computing needs.