Flash storage has already been around for a decade. During that time we have seen storage deployments transition from hybrid arrays to all flash systems as costs have come down.
Today, while All-Flash storage is the mainstream choice for most storage buyers, we are seeing the increasing adoption of Non-Volatile Memory Express (NVMe) storage/span>.
The ongoing transition to this new technology is expected to be very gradual - but inevitable. For the foreseeable future, we can expect to see the coexistence of hybrid, all flash and NVMe storage within the data center.
Legacy storage protocols are often considered to be the final barrier to unleashing the full potential of storage performance in enterprise data centers. But with NVMe, this bottleneck has been eliminated. NVMe is already successfully deployed in data storage in consumer devices, and now it is time for enterprises to also reap its benefits.
Enterprises demand extreme reliability, scalability and performance of their software-defined data centers – attributes which are well served by all-flash storage.
The road to transformed storage performance with NVMe
However the increasing prevalence of real-time workloads is now pushing the boundaries of all-flash storage, and this is where NVMe storage truly shines. For such workloads, storage and system I/O architectures need a complete overhaul to better support parallelism and multi-core configurations.
The adoption of NVMe in storage is expected to first start with NVMe-based storage arrays at the backend. Next will follow adoption of NVMe across the data center fabric (which needs support for Gen 5 or Gen 6 fabric along with NVMe over Fabric or similar). Finally, we will see the transition of storage-class memory with NVMe.
As we are still in the early days of deployment, it makes sense to use NVMe storage only for the most demanding of applications to justify the investment. The most natural adoption path is to use NVMe in a flash array cache to deliver the real-time performance that enterprises demand.
Solid state disk (SSD) caching involves temporarily storing data on flash memory in SSDs, which can be used for both data read and write functions.
In addition, using cache for data reads makes it possible to more quickly retrieve required data by placing previously requested data in cache- reducing bandwidth consumption as well as accelerating access time.
In the case of data writes, the cache temporarily stores data until the write operations are completed, boosting system performance.
The recently launched Fujitsu ETERNUS DX8900 S4 data center storage has been designed to significantly expand cache capacity to process huge amounts of data at high-speed.
It is equipped with a cache that can be expanded with NVMe flash SSDs located directly in the controller and it enhances standard SSD caching by using both a primary and a secondary cache.
The primary cache is configured with dynamic random access memory (DRAM) which helps to respond to server requests at very high speed. The secondary cache (or extreme cache), which significantly enlarges total cache capacity, makes use of NVMe drives to notably boost performance.
The power of NVMe - for a real business critical application
Let us step back and take a look at this from the perspective of how it affects businesses in the real world.
Consider an e-commerce consumer giant, handling millions of transactions every day for geographically distributed customers both in-store and online via the website or mobile application. This scenario has some unique features:
- Data access is very frequent, but the actual quantity of data involved for each customer transaction is usually small
- Customers are often very sensitive to response times, and it is often necessary to service many simultaneously, for example to support website access, billing or customer profile edits.
- Consequently, the demand for I/O access is considerable and primarily relates to random access.
The firm would have to allocate significant investments to purchase all new NVMe storage to meet all the business challenges. To truly benefit from the all-new NVMe storage, a parallel upgrade of the existing Networking architecture is also essential- which would imply an additional cost.
An alternative more straightforward and practical approach would be to use NVMe Cache in the existing all-flash storage. This will deliver a significant performance boost with minimal storage investment.
This approach would help improve response times in online processing, even when the customer count run into millions. NVMe cache ensures that critical tasks are categorized and handled by the extended system cache at the controller level before the data is passed to the drives in the storage array.
In today’s data age, NVMe is the new storage technology you want in your next flash storage purchase.
Adopting NVMe cache can be a first practical step in this quest.