Storage at the Speed of Ethernet
The future for video, IP and storage have at least one common foundation amongst them: Ethernet networking. Recent changes and future roadmaps for networking and storage systems are, without a doubt, about to alter the architectures of data-and-media storage as well as video transport in short order. We’ll look at some of these changes, both historically and in future tense, to provide some vision for what you should prepare for.
In the start-up days of Ethernet—around the early to mid-1980s—the strategy of “if you build it, they will come” propelled Ethernet’s development from a data-rate (ie., speed) perspective. In those days, if the storage community needed faster Ethernet, it was usually ready and waiting.
The novelty of PC computing extended to those who could afford the various components of the PC, including storage, network interfaces and memory. Storage networking was in its infancy and used only by those who needed shared data on a broader scale vs. in a PC workstationonly environment.
The cost to update or modify the infrastructure to support Ethernet was still relatively expensive; thus, major changes to networks were often delayed and were based upon the need and economics associated with those modifications.
The increases in speed for storage media weren’t occurring particularly fast either. Until the early 2000s, the significant changes in spinning media yielded only modest overall throughput improvements: faster and more reliable mechanics, rotational speed increases and reductions in latency. The big differences came in raw storage capacities, which increased many-fold from the early uses of rotating magnetic media. Ethernet, except for network transport, had little impact on storage systems as a whole.
Then, circa 2010, Flash memory as cache emerged to further improve the overall spinning-disk storage equation. As previous TV Technology articles have demonstrated, Flash dramatically improves storage system IOPS. And it doesn’t take a lot of Flash to make a significant difference in overall storage system improvements. Studies have shown that by adding as little Flash as 1 percent of the total storage capacity, in a tiered form, can yield increases in IOPS of 25 percent or more. Furthermore, as 3D NAND (Flash) was incorporated into SSDs, the cost/benefit ratio skyrocketed in many dimensions.
Fig. 1: Common Ethernet speed rates (link speeds) show current, in development and future standards with their dates of implementation. In the past, networking generally stayed ahead of other supporting system technologies, including storage. As storage systems have evolved, networking may now actually be lagging behind the storage (SSD) performance side, despite the fact more than a billion ports of Ethernet have shipped to enterprise, residential, data centers and industrial users.
Enterprise class SSDs, available today, can do sequential reads at around 2.8 GBps (22.4 Gbps). Essentially, this is faster than what a 10 GbE adaptor can support by a two times factor. It’s not difficult to see that in short order even 10 GbE network infrastructures will lag in performance to the level that storage throughput and/or system processing on network attached storage could be seriously affected; especially given the fact that 1 GbE has only recently peaked in terms of the total number of port shipments actually sold through this year.
As for 10 GbE, it is becoming cheaper and far more prevalent than it was two to three years ago. We’ve had 100 GbE since 2010, yet the 100 GbE growth rate remains much lower than 10 GbE is today. That said, 10 GbE is expected to reach saturation (peak port shipments) somewhere around 2018. By that time even faster mid-speeds (e.g., 25 GbE and 40 GbE) are expected to become common, and probably necessary for video systems where the transport shifts from SDI-for-video to IP-for-video (Fig. 1).
This gives rise to some serious considerations into how video and broadcast facilities will need to plan, network infrastructure wise, for the inevitable paradigm shift to IP.
NOT JUST FASTER
Putting the future for IP video aside, “Big Data” storage and the networking of that data is still predicted to increase by 50 percent a year. Once we seriously begin to pump video over IP, that number is surely to expand and possibly be incalculable by the 2017–2018 timeframe. Therefore, the perspective on network speed won’t be “how fast is your network,” but instead may be “how much more networking will you need?”
Not unlike what we’re finding for higher resolution video (UHD/4K)—“more, faster and better” pixels—is the question regarding storage networking solutions now “more, faster and broader” networks? Some, such as the Ethernet Alliance, believe that is the case and are in turn making a much stronger argument for addressing network improvement development.
The firestorm that is happening depicts an Ethernet ecosystem expanding by the second. Adding fuel to that fire, discussions are underway related to new direct connect 2.5 GbE and 5 GbE interfaces for HDDs. This, in part, appears necessary in order to address huge media storage systems, such as at Facebook, where billions of pictures and videos are now being housed on Ethernet-connected storage platforms. In addition, 2.5 Gig, 2.5 GigbaseT and 5 GigbaseT will allow existing Cat5e and Cat6 outlets to support new 802.11ac WiFi technologies; revitalizing cabling infrastructures in place worldwide.
The Ethernet roadmap has objectives for 400 GbE; with 100 Gb-single lane and Terabit Ethernet (TbE) as possible future speeds further on the horizon.
The reality of these higher-speed networks won’t become clear until the industry has reached a successful implementation of the 400 GbE and 100 Gb-single lane systems. Not forgetting that the products necessary to support IP-video must also include multiterabit fabrics in order to switch the 10-to-20 Gb full bandwidth (or even lightly compressed) signals for UHD/4K and beyond.
PARITY ACHIEVED
Some say that parity in SSDs and HDDs will be achieved somewhere in 2016; meaning the cost and capacities of the devices will be essentially equal. This will certainly change the landscape of storage systems forever going forward. As for networking, including Ethernet, this won’t be reaching that kind of parity for some time.
To the visionaries of tomorrow, keep these perspectives in mind as you plan your next “big” network system update; or as you consider the move towards an all-IP infrastructure.
Note: The author wishes to acknowledge appreciation to SNIA (www.snia.org) and the Ethernet Alliance (www.ethernetalliance.org) for providing background and statistics used in this article.
Karl Paulsen, CPBE, is a SMPTE Fellow and chief technology officer at Diversified Systems. Read more about this and other storage topics in his book “Moving Media Storage Technologies.” You can contact him atkpaulsen@divsystems.com.
Get the TV Tech Newsletter
The professional video industry's #1 source for news, trends and product and tech information. Sign up below.
Karl Paulsen recently retired as a CTO and has regularly contributed to TV Tech on topics related to media, networking, workflow, cloud and systemization for the media and entertainment industry. He is a SMPTE Fellow with more than 50 years of engineering and managerial experience in commercial TV and radio broadcasting. For over 25-years he has continually featured topics in TV Tech magazine—penning the magazine’s Storage and Media Technologies and its Cloudspotter’s Journal columns.