Questions from Storage Trends Webinar Answered, Part One
It was a great pleasure for Rick Kutcipal, board director, SCSI Trade Association (STA), to welcome Jeff Janukowicz, Research vice president at IDC and Chris Preimesberger, former editor-in-chief of eWeek, in a roundtable talk to discuss prominent data storage technologies shaping the market. If you missed this webinar titled “Storage Trends for 2021 and Beyond,” it’s now available on demand here.
The well-attended event generated a lot of questions! So many in fact, we’re authoring a two-part blog series with the answers. In part one, we recap the questions that were asked and answered during the webcast, but since we ran out of time to answer them all, please watch for part two when we tackle the rest.
Q1. How far along is 24G in development?
A1. Rick: The specification is done and most of the major players are investing in it today. Products have been announced and we’re also expecting to see server shipments in 2022. STA has a plugfest scheduled for July 6, 2021. It’s a busy time and everybody’s pretty excited about it!
Q2. What’s after 24G SAS?
A2. Rick: Naturally, one would think it would be a 48G speed bump, but it’s not clear that’s necessary. There’s still a lot of room for innovation within the SCSI stack, not just in the physical layer. The physical layer is the one that people can relate to and think “oh, it’s faster.” Keep in mind that there are a lot of features and functionality that can be added on top of that physical layer. The layered architecture of the SCSI stack, enables changes whether it’s at the protocol layer or another higher layer, without impacting the physical layer. These are happening real time and STA is having T10 technical committee meetings on a regular basis, and innovations are in the works.
Q3. Where does NVMe HDD and 25G ethernet HDD fit in?
A3. Jeff: Generally speaking, it’s still unclear how that’s going to evolve. As we look out over time, in the enterprise market on the SSD side, clearly, we’re seeing NVMe move into the majority of the shipments and SSDs are growing as a percentage of the overall unit shipments and petabytes. However, right now we’re seeing a mix of technologies that are used within a storage array or in an enterprise system. And clearly, they are SAS-based SSDs and HDDs. And with that transition to more SSDs, it’s sort of a natural question to say, “hey, what about putting the NVMe interface on HDDs?” Now you obviously don’t necessarily need it for all the performance reasons or the optimizations around non-volatile media, which is why NVMe was introduced, but there are some initiatives, and these could help bring some cost savings and further system optimizations to the industry. There are some things underway from OCP in terms of looking at NVMe based HDDs, but they’re still relatively early on at least from my perspective in terms of their development. But there are definitely some activities underway that are looking at the technology.
Rick: From my perspective, I’m seeing a surge in NVMe HDD work within OCP. My concern with NVMe HDDs is the amount of standards work that still has to be done to make them work in an enterprise environment. I think people forget it’s not just taking some media and putting an NVMe interface in front of it. How do all the drive inquiries get mapped to NVMe? How do you manage enterprise large scale spin up? I think it’s an exciting time. I think there are a lot of good possibilities, but the amount of work that’s needed can be underestimated sometimes.
Q4. Could you discuss the adoption of SAS SATA and NVMe in all flash arrays?
A4. Jeff: IDC has seen a lot of investment in terms of all flash arrays. And we’ve seen pretty rapid growth over the last couple years. In 2020, about 40% of the spending on external storage was on all flash arrays. And the reality is if you look at that today, the vast majority of those are really still built upon SAS-based SSDs. There have been some announcements from a lot of the large storage providers around NVMe-based arrays, whether it’s Dell EMC, Netapp, Pure Storage, IBM, etc. Today, these solutions have already started to become available in the market. And we do see NVMe AFA’s as a very high growth category over the next few years, but right now they’re still targeted primarily at a lot of the higher end and more performance-oriented types of applications. We’re really just starting to see them move down into the more mainstream portion of the all flash array market. Which from IDC’s perspective, if it was 40% last year, we see it growing as an overall category to about 50% of the overall spend on external storage by 2023. So clearly there is a lot going on in this market as well.
Rick: My questions in regards to NVMe and all flash arrays is always about scalability. I know there’s a lot of work going on regarding NVMe over fabrics, but if you go back and look at the amount of computational resources, memory and system resources that it takes to scale these things, there’s still some pretty big challenges ahead. I’m not saying it’s not going to happen, but of course the ecosystem, has solved hard problems in the past.
Q5. How do you differentiate between M.2 SSDs and NVMe in client system deployments?
A5. Rick: The SOCs or the controllers on these devices are very different. There are enterprise class M.2 drives, so the form factor doesn’t necessarily preclude it from fitting into one of these categories. While M.2 is more designed to the client, it’s not a hard and fast thing. Typically, it’s the traditional 2.5.
Jeff: Rick, you’re pretty much spot on. There are some differences at the SOC level and design level such as power fail protection. But there does tend to be a different firmware load a lot of times for the enterprise class drives. There can also be some differences in terms of the endurance in how those drives are designed. But if the question is about form factors, we really are at an interesting point for the industry, because historically it has always been dictated by HDD form factors. But as flash has grown, we’ve seen a lot of new form factors. M.2 is obviously one that was originally designed for some of the client market, and has now found its way into a lot of enterprise applications. E1 short is a slight variant of M.2 but is on the roadmap to be more enterprise optimized form factor. But we also see some other ones out there like E1 long, which is a longer version of E1.S. There’s also U.3 and others which are pretty interesting in terms of ways to optimize around some of the new storage media, i.e., SSDs and solid state.
Q6. Is the NVMe takeover sooner than 3-5 years?
A6. Rick: That’s a very logical question. People that aren’t in the ecosystem day-to-day might not be seeing the 24G SAS adoption. Right now, there’s a lot of investments at the system and sub-system level. For 24G SAS there are multiple adapter vendors, same as there has been in the past for 12G SAS. And from the media side, there are numerous drive vendors sampling 24G SAD drives today, and one has been announced. I think some people are going to be shocked of the 24G adoption, and that’s going to start coming to light at STA’s next plugfest, with some big demos and press announcements as products get ready to launch. So, I guess I would, say stay tuned for that one because I think people, some people, are going to be pretty surprised.