Preparing For A Superfast Future With NVMe

by George Chacko    Mar 16, 2017

speed

Flash storage has taken over the data center. which has proven to be the most disruptive storage technology of the past few years. Especially in the case when we are dealing with delay sensitive applications, high-definition video, mobile computing. Flash storage is significantly faster, with lower latencies than mechanical spinning disk drives. Software and hardware vendors have leveraged flash to increase performance throughout the data center.

Users expect their applications to respond immediately and flash storage helps deliver this level of performance. In fact, moving to flash storage in the data center offers several advantages, such as:

- Less power consumption—some vendors claim 600x better power efficiency than HDDs- Less heat production than traditional spinning drives- Less space required- Lower latency—some vendors claim a 40x reduction compared with HDDs- Fast return on investment- Better performing applications

Don’t forget about your network

Although you may have invested in flash storage for all the anticipated benefits, you may not be getting everything that you expected from your investment. Increased traffic between storage and servers, for example, can quickly oversubscribe your existing network links. In some cases, the transition to flash pushes the data access bottleneck away from the storage devices and toward the network.

This situation can lead to extended ROI cycles for your flash storage investments or can slow your flash implementation altogether. It is critical to understand your performance bottlenecks and potential challenges as part of such investments.

Two key benefits that can be expected from flash include better performance/throughput and lower latency. But without a balanced infrastructure of compute, storage and networking elements. you may be minimising the flash technology’s potential. With aggressive deployment of flash across the industry to increase performance, this problem will only get worse for many companies.

What options do you have?

 To deliver good performance levels with new flash devices, the industry is embracing a new flash storage protocol called NVMe (Non Volatile Memory Express), a storage protocol built explicitly for flash and designed to improve data and latency sensitive applications. NVMe commands to travel natively across an existing network and this extendsthe protocol over large storage fabrics.

Although NVMe typically serves in direct server attached storage today, there’s a growing need to extend this low latency protocol over storage fabrics, where shared storage solutions can benefit. In addition, it promises to increase scaling to as many as 1,000 shared storage devices.

To add strength to the protocol, a new industry standard – NVMe over Fabrics – was announced in June last year. It allows NVMe commands to travel natively across an existing network, enabling you to now extend the protocol over large storage fabrics.

Owing to its nature, NVMe can run across a variety of fabrics, including Fibre Channel and Ethernet. It allows you to use your existing network architecture by standardising on a common abstraction layer that can operate over multiple network fabric types. 40Gbe with iSCSI, file-based network file system (NFS) or server message block (SMB) and 16G Fibre Channel will all be a network bottleneck when compared with NVMe over fabrics. It’s similar to the lack of performance you would get from driving a high performance road car on a dirt road.

 

When datacenters deploy new flash arrays it is typically with Gen 5/6 Fibre Channel as the preferred network. As these flash storage devices transition to NVMe, IT leaders can seamlessly and transparently deploy new NVMe-based ultra-fast applications without upgrading their FC network.

 

What does this mean for your business?

Ultra-fast data transfers is common amongst most businesses today across large-scale networks for business critical applications. NVMe over Fabrics dramatically reduces latency and eliminates the need for SCSI translation, by directly transferring NVMe commands and structures from end to end, thereby making applications run faster or scale better.

A primary goal of NVMe is to reduce protocol overheads seen in traditional (Fibre Channel and Ethernet) encapsulation techniques and thus reduce latency and increase IOPs between the host and the target storage device. NVMe over Fibre Channel (FC-NVMe) works with flash storage to ensure you get the performance and low latency that you originally wanted, but it also provides the additional reliability and performance of a Fibre Channel network.

Running NVMe natively across a Fibre Channel network provides benefits of flash storage, maintains flash simplicity and efficiency, while eliminating the need for translation. This will lead to higher application performance, more data storage, better analytics and more personalisation of information.

For many, NVMe over Fabrics represents a high-performance solution with low latency. A primary advantage is the ability to add sustainable scale without affecting performance.

By delivering low latency for all-flash arrays, NVMe over Fabrics is emerging as a promising solution for large-scale enterprises and SAN infrastructure. Even if you’re not ready to make the move today, you still might want to prepare with Gen6 Fibre Channel solutions as you move toward NVMe over time.

Preparing for the future

NVMe and Fibre Channels, allows you to extend the benefits of flash in latency and performance at the highest Fibre Channel speeds. By investing in hardware that supports NVMe today, you can ensure your network and storage are optimised for whatever’s coming next. This will enable you to support enterprise data centers, mobile computing, high-performance computing, relational databases and other uses of the technology.

As applications continue to advance and virtualization permeates the data center, flash storage will grow right along with them. The unprecedented speed and rapidly increasing cost-effectiveness of flash-based products are dramatically accelerating datacenter transformation. One thing that is very important to understand is how much bandwidth response your applications require and what application response time growth you foresee.

Rethinking your current data center environment to accommodate flash and the applications that require it, will most likely bring you to a discussion about NVMe and NVMe over Fabrics. Understanding how communication occurs among storage, compute and network resources will greatly affect your success. As the next evolutionary step for your storage network, you can you can implement the best solution for your business, using NVMe.

[The author is Principal Systems Engineer & Lead Technical Consultant, Brocade India]

[Disclaimer: The views expressed in this article are solely those of the authors and do not necessarily represent or reflect the views of Trivone Media Network's or that of CXOToday's.]