Storage Developer Conference - #150: Tiered Storage Deployments with 24G SAS

Episode Date: August 3, 2021

Hello, everybody. Mark Carlson here, SNEA Technical Council Co-Chair. Welcome to the SDC Podcast. Every week, the SDC Podcast presents important technical topics to the storage developer community. Each episode is hand-selected by the SNEA Technical Council from the presentations at our annual Storage Developer Conference. The link to the slides is available in the show notes at snea.org slash podcasts. You are listening to SDC Podcast, Episode 150. Hello, and welcome to the Storage Developer Conference 2020. I'm Jeremiah Tussey, Vice President of the SCSI Trade Association. I'm also the Alliances Manager of Product Marketing of the Data Center Solutions Business Unit at Microchip. Today, I'm going to talk about tiered storage deployments with 24G SaaS.

So as some background here, SaaS is serial test guzzy, which is the only storage interface that embraces both high performance and high reliability, as well as providing native compatibility with low cost per gigabyte SATA drives or devices. Overall, this capability allows SAS to span a variety of storage environments, including tiered storage solutions, which is part of our topic here today. This deployment is at a degree of large scale, utilizing tens of thousands of hard drives and SSDs, so quite fast in its deployment. So overall, we'll focus on some main objective areas, highlighting the latest features of SAS, the benefits of this tiered storage, as well as how to take advantage of the 24 gig SAS

for this deployment. As an overview of our agenda we'll focus on today, I want to touch base on some background. So given some context using some enterprise storage market data to kind of give an idea of why SaaS is where it is today and where it's going in the future and why there's a need for this deployment and tiered architectures. And then to support that deployment, we'll look at the 24 gig SaaS enhancements and where the specification developments have been driving towards to support that and get more into the application side of things. So we'll take a look at some specific example cases, some overview of the marketscape and kind of touch base on some areas that SAS and SATA are deployed today or where they could be deployed. There's lots of uses and benefits to using SAS that span across the performance and capacity spectrums.

And then dive a little bit more deeper into the storage tiers and why we need to do different storage tiers with SAS deployment. And finally, I'll wrap up with some key takeaways to give you some take-home notes and allow the opportunity through this forum to ask any questions, and we'll provide answers for the Scutzi Trade Association to our best ability or as your questions come up throughout this event. Okay, so jumping into the market area, the market data, I wanted to kind of give you a view and some background as to why we're using SAS today and why we'll continue to use SaaS and SATA for that matter. So what I'm

showing here is a market view from analyst firm TrendFocus as recent as August 2020. And what we have showing is all enterprise storage exabytes from last year going into a forecast look out to 2024. And so what this includes is SAS and SATA hard drives and SSDs, as well as MME SSDs, so all the primary storage that you see in enterprise deployment. And as you can see for the TRIN, SAS is the connected storage that continues some very strong growth. This is a concept that's been in play for a while, and it just as an example, the near-lying hard drives that have been shipped in the last year was as recent as this previous quarter were 170 exabytes. And this will span out to over 2,000 exabytes out in the year 2024.

So very expansive growth. And in comparison, the SSDs over the last quarter were at 20 exabytes. So quite a difference in scale and deployment. So overall this shows where hard drives are still majority of capacity shipped upwards towards about 90% of the form factors that are deployed. So basically SAS is not going away anytime soon. It's the favored storage media of choice for those capacity deployments where dollars per gigabyte matter, as well as existing infrastructure and the need to be consistent with your deployments and getting the capacity growth that are supported in the market today.

As we zoom in, so this is the section that was at the bottom of the previous slide. This is a zoom in on the SSD portion. So again, from Allen's firm, trend focus coming from August, this gives you a view of SSDs only and where we're focused on SATA, SAS, and PCIe or NVMe SSDs in the enterprise storage space. So here you'll notice SATA SSDs are more of a flat market while SAS is still slowly growing. But it still overall represents about 26% of the market out in the 2024 timeframe. So not insignificant, but a bit more muted because of the differing needs for the deployment of storage. So enterprise storage supporting NVMe as a bigger growth factor going forward. And just to highlight that that tipping point did already happen.

It happened last year where NVMe outshipped SAS and SETA combined by about two exabytes for that year. But, you know, that doesn't take away from the slow growth that's still there in the SAS SETA ecosystem, which is supported by SSDs to some degree, even a larger degree by hard drives. And NVMe will be shipped into specific markets as needed by the server side, hyperscale especially. And SAS and SATA is more service in varying markets such as enterprise and data center applications. So overall, as a takeaway from this, you'll see this continuing growth, especially as the CPU platforms go through their transitions in the coming years to next generation speeds and feeds. That's where more NVMe direct connectivity will happen,

which also drives the need for more external storage, which will be attached to SAS, SATA, interconnectivity, HBAs, expanders, and thus SAS and SATA SSDs and hard drives. So a continuing good outlook for all of these different storage solutions that are provided today. And one last note here, too. One of the big differences between the hard drive deployments and the SSD deployments, if you look at the NAND flash that's shipped in the market today,

shipped in the world into all the various applications, if you were to take all of that and combine it together, the amount of flash that's available in the world today is it's still nowhere near to a scale that could be a replacement for hard drives so you know the you to get that capacity you're going to need hard drive still late there's still the the the primary shipping form factor today. Okay, so getting into more of the SAS implementations of new media innovation. So one of the benefits of SAS is it allows the adoption of new innovations and these different media interfaces so we could focus on the ever-increasing IOPS need, capacity needs,

and provide the lowest cost per gigabyte with various solutions. A lot of these focuses are on the hard drive side. There are some on the SSD side. But overall, the end contribution of these different technologies is adopted by SAS architecture and specifications. So what I'll do here is highlight some of these

that have been coming out over the years and some of the newer ones, and see how these are being adopted and implemented by SAS and SATA specifications and technologies. So one of the first ones that has been out for a while is multi-link SAS, which is an implementation of a wide port interface on SAS drives. So what this does is allow for performance having multiple channels to increase the bandwidth from a hard drive or SSD if a user so

chooses to implement that on a drive, as well as implemented on the server or storage side to interconnect that wider bandwidth capability. So that's just a mechanism to take advantage of the inherent bandwidth capabilities of the protocol and put it into hardware. And right behind that, we have storage intelligence, which is more targeted for SSD implementations where you're streamlining the performance. You got dedicated SSD commands. So it gives you a way to optimize the SSD functionality including background garbage collection and streams providing streaming capabilities, which is important in some

applications that are deployed with SASES and SATA today. Switching back over to some hard drive technologies that have been developed over the years, TDMR, which is two-dimensional magnetic recording, is one of those enhancements on hard drives that allow for faster write performance, contributing to HDD connectivity. And that, together with some of the more recent introductions of HAMR and MAMR, which is heat-assisted magnetic recording and microwave- assisted magnetic recording, those are two technologies that are used to enhance

the capacity of drives. So that's where our drives will be increasing the bits per inch. And this is another case for contributing to HDD connectivity in SAS environments. And then not to be outdone, there's also been this introduction for a number of years of SMR, which is single magnetic recording. So this is an implementation of narrowing the tracks on the platters of a hard drive. So there's a little bit more complexity to dealing with this when it comes to reading and writing and how you have to streamline things. So it kind of gives more of a, I don't want to say tape functionality, but more of a specific functionality that contributes to more of a sequential writing mode on the drives. And this requires specific control and monitoring from the host so that this is implemented in a way that benefits the end user in providing the increase in capacity, which can be upwards to 20,

maybe even 30% increase in capacity in those cases on the same type of media that you see with standard shipping air conventional magnetic recording drives today. So this is something that's supported by the zone block command set specification that is part of the SAS role. So SAS and SATA development activities and the standards have started off with that implementation, which helps to take advantage with some of the other enhancements that have come, like helium. So there's been helium deployment in drives today that increases the effectiveness inside the drives

where you can lower the heat and the friction generated in drives. So all of these things combined are contributing to more and more capacity because that's the demand for hard drives today. And one of the most recent incarnations of SMR is hybrid SMR, hybrid single-banded recording, which is now part of a new ZBC2 standardization. And this is more focused on the flexibility of using a single mandate recording, where you can have the same type of drive be deployed with either 100% conventional or 100% SMR or some combination of the two to have that flexibility for the demand for increased data versus standard performance and access by certain server environments. So as you can see, more and more flexibility provided by a SaaS ecosystem. Now, some of the more recent introductions, you can say, are multi-actuator.

There's been a lot of buzz and talk about multi-actuator or dual actuator by several vendors in the market today. And this is yet again another hard drive technology where we're focused on performance, but not just a general performance, but more of a IOPs per terabyte metric. So when you have a deployment with all this capacity, specifically in a hard drive media, you have what can be a limitation on the interface from that media, that spinning media, to the host. So this contributes to the increase of that IOPs for that growing capacity to get that up to the host, and so you can write and read to and from that drive to scale with the ever-increasing capacity that is being deployed with these hard drives. So how this is supported in SAS today is

with multiple LUN connections. So implementing various LUNs within the SAS architecture you can access those those individual actuators on the drive as like their individual drives, but have the benefit of increased layouts. And this technology has a future going forward to adapt 24 gig SAS or other requirements that don't necessarily need to take advantage of multiple ones. So look for those types of deployments going forward, but all adopted by SAS, and consequently some possible SATA connectivity options there. And last, but I want to say least, is QLC,

which is Quad Level Cell, NAND. So in the Flash environment, this is one of the more recent introductions that allows the growth in capacity on the SSD side. Now, there's some tradeoffs with QLC in that this increase in capacity and this density within the NAND architecture requires probably a slower access time than is typical with NAND. So what you see with this type of NAND deployment is something that is more akin to SMR in its deployment. You have to have a little bit slower access so that you can have a bigger capacity. And so you want to treat these SSDs in a little bit different manner and try to streamline the access to those SSDs to make a more efficient and long life of demand over time. So some of the standardization that's going on there has been adopted in the NVME side

of things with zone name spaces, and that standardization is aligning with VDC, which is the implementation of the single-magnet recording. So similarities in how to access these types of media, even though it's different technologies, hard drive versus SSDs. So now that I've spent some time on the background of why and what we're adopting on the different media front, let me go a little bit more into the standardization and where we're focused on in the SAS side of things, the SAS world and those standards. So here we have multiple layers that are developed to support SAS.

So starting off with a physical layer, where we've introduced these standards, we're evolving with PHY technologies and the processor interface transitions to support different data rates and capabilities. One of the earlier introductions was 6 gigabit per second rate for SAS, supporting zoning and eventually adding some managed cable capabilities, power management,

and then adopting some of the needs of the enterprise ecosystem. So all these physical layers are compatible with SAS, as well as have an open compatibility with with SATA serial ATA at a physical level and so this these are designed to support those different types of PHY interconnects and enhance the the signal of the PHY itself so over time we kept adding more and more features to enhance the channel for each upgrade to the physical data rate. And in our current incarnation, we are at 22.5 gigabaud. And at that level, we've added some forward error correction, changed the encoding to support different insertion loss specifications, and it added some different connectivity options

overall to support the ever-increasing speed requirements for each step, but keeping the reliability of the data rate and the signal integrity of that connection. So that's the focus of the physical layer. Now, if we go one step lower, we start looking at the protocol layer where we're supporting the packets, the communication of the SCSI command set over that SAS physical layer. So here we have some implementations to adopt to the transmitter level, so we call it transmitter training, making adaptations for optical cabling. And once you're up and running, start looking at the persistent connections

and power primitives where you're supporting different power modes. So overall, you want to adopt to the different connections that are happening in the real-life environment. And some of the more recent introductions in SAS 4 have to do with increasing fairness across the span of SSDs versus hard drives, which can be SAS or SATA. So you have different things running at different data rates, and you want to make sure things

run at a fair pace and communicate properly. So we've added even more visibility and management to support those different connectivity options. And last but not least, also acting on the PHI side, adding an active PHI transmitter adjustment, we call it APTA, which allows for dynamic adaptation of the channels, considering the varying environments that can be in place, whether it's temperature, voltage, variations, and whatnot. There's different effects that can be affecting as low as the silicon level, and these abilities allow you to adopt to those in a live manner. And then notwithstanding, we look forward looking, we're looking to adopt even more expensive flexibility and connectivity for retimers, wide ports, interleaving, you know, just make sure we have a good spectrum for SAS as a whole.

So beyond those two main physical and protocol layers that comprise SAS, which with the current generation 24G SAS being made of SAS 4 and SPL 4, you have some enhancements beyond that that are part of the SCSI command layer. So with the SCSI command layer, you have a collection of specs specified to SCSI itself, which is the foundation of SCSI. And here we have the SCSI primary command SPC, which is a mandatory and optional command level of specifications for all the end devices. You got SCSI block command SPC, which is a command set extension that uses SCSI in storage environments and then going into a SCSI architectural model which is a SAM specification this defines reference models of common behaviors for all SCSI devices on the management side you have

SESS which is SCSI enclosure services and here we we document commands or parameters to manage non SCSI elements installed in the enclosure, such as fans, power supplies, temperature sensors, you name it. There's different components to enclosures that are required to keep the health up to par. And the SCSI enclosure services allow for communication to a lot of these other components through some kind of baseboard management interface or enclosure management built into the SAS infrastructure itself. There's various ways to implement that,

and SAS is the structure for that. And then last but not least in this layer is SAT, which is the SCSI to ATA translation. And this defines the protocol requirements that bridges between the SCSI and ATA devices, allowing for SCSI components to interoperate with devices like SATA hard drives and SSDs. So that's what allows us to have this vast flexibility of varying components. On a different level here we have the ZBC command sets. So ZBC as I mentioned earlier is zone block commands and this defines the model and command set extensions to facilitate the operation of zone block

devices like SMR drives and that expands into ZBC2, which has a different focus towards dynamic zones, which is part of the hybrid SMR, and new sanitized rules to take advantage of with the SMR deployments. So this is all the infrastructure that's supporting the SMR and will be looked at for QLC deployments if they're deployed in SAS setting environments as well. Moving on. Let me change gears here and look at some of the data center storage applications that you see with SAS and SATA. Here are some common data center applications that are just a sampling of what can be deployed or what's out there today. There are obviously other applications and environments that can certainly fit into this slide but you know just want to focus in on a few here that the that give some context and what we're showing here is the x-axis

representing a relative performance across these these varying applications while the y-axis is more of the application intended for reads or write type of operation, whether it's read-intensive or write-intensive. So different levels of performance metrics. And the size of the application bubbles are meant to represent the general range across each access with large with the larger bubbles being a greater range for reads and writes as well as performance demands so kind of give you a different basis of needs for the different metrics here so as we look at the different bubbles here, we can pick out a few of these and think about the applications here.

So if we were to start at the top of the spectrum with financial trading, you have some examples like Robinhood or Webull, which are app-based trading platforms. And what the requirements there are needed is you need very fast access to make your trades in the market. You need to get your order in. You need to specify your price, your limits, and that needs to execute very quickly because you're competing with a bunch of other trades in the market. So the platform supporting that can be very write-intensive and very high performance. That would probably be serviced by a hot-tier implementation

using the highest-speed SSDs, whether it's SAS or NVMeet, for that matter. But this is an area where the demand is for high-speed general deployment of SAS architecture as an example. Now, as we dig into some of the other applications, you can look at AI here over towards the left. An example of deployment there would be someone such as Palantir. It's been in the news more lately. They have business and government situational awareness and decision analysis that they're implementing with their technology, basically focusing on surveillance tools and some

applications stating that they're hunting for terrorists, rogue traders, a wide variety of applications that might sound a little bit big brotherish, but there's a need for these types of analysis requirements, and they're providing that service. And, you know, you have different levels of light intensiveness and read intensiveness that is required at a fairly high performance level, which could be serviced by SAS technology. And then again, going back up the spectrum on the performance side, you have gaming and VR, a very popular area of deployment, especially on mobile gaming, where you have new games

such as Eve Echoes that spans across the spectrum of multiplayer, massive deployment, lots of people around the world getting into an environment, competing or working together in a gaming environment that takes a lot of resources in the background to make it all happen. And as far out as augmented reality, some might recall and maybe some still use Pokemon Go, where you have a virtual reality a virtual or augmented reality implementation on your mobile device as you go around finding some mysterious hidden Pokemon figures that all requires

data to be sent up to the cloud and processed and fed back to your mobile device for the end user interaction. And then you can go into a crossover of some of these applications even. So if you were to look at AI and business analytics, for example, one example of an implementation that would span across those areas would be the Event Horizon Telescope project. I don't know who all has heard of this project, but it's a pretty massive undertaking where they collected 4.5 petabytes of data over five days, mapping a photo of a black hole.

So quite an extensive project that required some artificial intelligence to stitch all these photos together and the analysis to implement all this together. And this was actually deployed with SAS technology where WD 8 terabyte helium hard drives were highlighted as being used because SSD cost and capacity limitations at the time. So just another example of SAS in a data center storage application. And, you know, I can go into more details on all these different ones, but let's just say I'll wrap it up with one other area that is probably kind of prominent right now is content delivery.

With people staying at home more often, there's more streaming of movies, television, especially as people do Cut the Cord, for example. And that content delivery is supported by Netflix, Hulu, Disney+, Amazon Prime, many others. And that's a big segment that requires some performance, not really white intensive, but definitely highly read intensive, and streamed in an efficient manner to multiple households and multiple users per household, as is the case in my household.

You've got multiple streams at a time. So there's quite a few different uses that are supported by SAS and SATA. It's very, very flexible and spans a wide variety of applications. And as we look at it a little bit further just to give some context. You know, in the current world we're in today, data is everywhere, and it's in every way. For example, Internet of Things. As you saw in the previous phase, Internet of Things is one of those application bubbles that is pretty large in size and is supported in a large way

by SAS SADA architecture, or can be going forward. It requires quite a bit of capacity if you look into the details. So what is Internet of Things? I mean, if you think about it, you've got your Samsung smart things. You've got Alexa. You've got your doorbells, your video doorbells. You've got your digital smart thermostats. You've got light switches, surveillance security systems. You've got a lot of things that are going digital and more user controlled and are connected to the cloud. And just as an example, I recently connected my fireplace to Electa. So for convenience, I could say,

turn on my fireplace and I don't have to flip a switch anymore. I can still flip a switch if I really want to, but it's just one of those cool factors, convenience factors, that's being adopted all over the world in a vast way. But behind all of this, you need to have some kind of storage infrastructure. So to give some context here, if you were to look back in 2018, there was some interesting data that came out. And it talked about how 7 billion devices were connected to the

internet. That was in 2018. Now just think about that now, what that means. So if you look forward, now they're saying there's 41.6 billion connected IoT devices generating 79.4 zettabytes of data looking forward in the 2025 timeframe. That's IDC's stated metric from last year, but it just kind of gives you an idea of the scale of what's going on behind all of this technology deployment and that's notwithstanding some of our more favorite platforms that a lot of people are using today especially as as your home more you know there's there's a lots of dynamics in the world today so you have you know Facebook and Instagram where you have every family video and photo that's being stored.

That's everybody's gateway into the world. They're friends and users. A lot of people are communicating through Facebook, especially in these COVID times where people are more at home. A lot of our communication is virtual. And those are some of the platforms that are supporting that. And then YouTube, where you have every tutorial and experiment that's recorded and posted. And they even branched out into television deployment as well.

But YouTube is a massive platform supported by users today, as well as Twitter, lots of tweets back and forth. Everybody hears about some of these tweets from more popular and famous people that makes headlines sometimes. But the platform is there to allow people to do their little blurb out into the world. And again, you need storage in the background to make it all happen. And then, you know, LinkedIn, Snapchat. LinkedIn is more of a business interface application, kind of like a Facebook, but for the business world. Snapchat, where you can do your little clip-its of videos or pictures more privately or just for fun to your friends and family.

WhatsApp is more of a communication-type platform, so you can do kind of a text messaging type of implementation over the cloud where you can send pictures and videos or communicate using a video or voice-over IP type of call with friends. We have friends in Germany that we communicate with using WhatsApp. And those are just some of the applications and not to discount some of the other areas like Pinterest where you're pinning your interests of, you know, your favorite recipes, your favorite character, or there's movie stars, something like that.

And then TikTok. I mean, look at TikTok. You got people that are recording every single dance move and posting on TikTok. They're a little bit more into headlines these days, and who knows what their outcome will be. But the overall underlying message here is these are all using data. a little bit more into headlines these days and who knows what their outcome will be, but the overall underlying message here

is these are all using data. And if you were to put a metric on this, you can look at, for example, YouTube. You have an estimated 500 hours of video being uploaded per minute.

That's just a staggering amount of video data and on the tick-tock site 1 million tick-tock videos watched every day so ever increasing demand for storage infrastructure now just to give an idea of how this is supported from a tiered storage architecture, this is a social media example. Some people's favorite social media platform, not that you can't tell what this might be, but let's break it down and take a look at what's involved here. So when you look at this social media interface, what do you see? You see some stories. You see some rooms that are shown up.

These are areas that you have user generated photo video content that is only available for 24 hours and in the story section rooms you can instantly video chat with friends. This is all data that you need that needs quick access so what you're going to have here is a hot tier behind that because that's going to be your more of your instant access interface now some of the areas you have highlighted in yellow here are more of a i would call a warm tier this is where you're navigating accessing the community which includes links to groups that the users might be in, as well as other services and suggested pages that are popping up there.

You have in the center there ads and promotions, which uses the user data, whether it's their location, demographics, their profile, their friends, their likes, to display relevant products and services. And then the chat. So you need somewhat quick access on the chat because you're tuning in back and forth, but it's just a text chat. So not very high bandwidth. So this can be supported by a warm tier type of application. And then you have towards the bottom there, highlight in a blue, more of a cold tier. Granted, this area might be more of a hybrid because some of the more recent feeds

are going to be probably on a quicker tier. But over time, none of this data goes away. It just gets relegated to a colder storage. So then anyone can go back in history and look at some older posts. And this is a dynamically updated list of posts from your friends and groups and news or whatever else that

Will be available indefinitely. So you you you don't want to put it on your most expensive storage You start putting on your your cheapest storage, which is you know A cold storage implementation using SAS or SATA hard drives, probably SMR type drives. So as I discussed for those different areas of that social media interface, this is what's behind it. So you'll start off with a web server of some sort that provides the gateway into the infrastructure, which is just going to be a standard off-the-shelf, probably low capacity, not necessarily needs to be high performance. It just needs to be the gateway into all the data. So from there, you'll get into your database for your user data, which is built on a warm tier using SAS hard drives

or SSDs with arrayed implementation. So you're looking to provide some medium capacity, some high IOPS at times, but it's not a constant high-performance requirement. And then the object store, which is the cold storage. This is where you're going to store all your data that's aging over time. This is going to be supported with large deployments of SATA devices where you have very high capacity. The requirements for performance are low. When someone goes and looks at someone's older posts,

you might notice it takes a little bit longer to load it, not that long, but it's because it's being stored on a not-so-hot tier somewhere out there in the cloud. And then you have your cache, which is your hot tier. This is your media access for applications and user data, and it's going to be implemented on a couple of SSDs. You'll have lots of RAM, lots of memory, so that you have that very high access speed. But your capacity is going to be limited to maintain a balance on your cost. And these are the different implementations overall in the social media application deployment. But beyond that, what about analytics?

With a data set of millions of users on these types of platforms, what can you do? I mean, you have access to user location, demographics, their friends, their likes, and all of this contributes to, you know, providing tailored ads so you can get them to click on something and monetize the platform because, you know, these platforms are free for the users to access usually. So you need to make money somehow, and that's understandable. But how do you do that? How do you engage them to make sure that they're going to respond and take advantage of those things that are monetized? Not only that, but streamline the user interface so that you can use those search metrics to find new friend suggestions. Connect them with other people in the world, whether they're people they know or not, just maybe people with similar interests.

And more than that, recommend services or suggest certain pages that you think they'll like. So there's a lot of that analytics and tailoring that's done in the background. So what does that look like from a hardware perspective? perspective here I've put together a diagram that kind of gives an outline line and map these to the previous slides where you show the different portions of hot cold and warm tears and again you got your web server which is the interface to the world going through that you got your hot tier which is your memory cache so instant access directly attached to the web. Going through that, you got your hot tier, which is your memory cache, so instant access directly attached to the web server. But you're also directly attached to your warm tier. And this is where you have your database, and

that database references what's stored in your object storage, where your videos, your old videos, your old photos, are going to be stored in that cold storage and accessed through that database server. But off to the side, you notice the analytics, an analytics type server. So I mentioned analytics. You're going to have another service out there that's probing this data, gathering this data, you know, monitoring this data, gathering this data, monitoring this data. And that's what's going to be used to analyze and put all the pieces of the puzzle together

and use something like a feeder server or something where all the outside ads of different companies, different groups are stored in that. And it goes off and grabs those to feed into suggestions back to the user. So just how the whole implementation comes together to provide the user experience and done in a tiered manner so that you can have the most

efficient deployment of storage. So what's this mean? What are these tiers about? When we look at these tiers, I mentioned three different tiers. I mentioned the hot, I mentioned the warm, and I mentioned the cold. You look at the hot tier. You're basically a server that may be directly attached to your highest performing type of media, which in this case, today's technology, 24 gig SAS SSDs. In some cases, maybe NVMe SSDs. But let's look at the SAS case. So you have 24 gig SAS attached to a controller inside the server as one example, or you could be attached to a storage array enclosure that would be SAS or SATA SSDs in the form of an all-flash array that is attached

to the server. So those are ways to implement the hot tier. And this services applications, such as what I mentioned before, like financial trading, gaming, virtual reality, augmented reality, the business transactions and AI, where you need some quick access metrics, low latency, high IOPS. You want to focus on SSD connectivity to the host and to the server. So the specific storage requirements for the hot tier is performance, gaining that IOPS that you need for your application without spending a whole lot. The deployment is typically a couple of SSDs all the way up to maybe 16 or 24 SSDs to get a balance of some capacity, but maintaining a performance metric. And in today's applications, you can

see direct attach up to 24 gig SAS SSDs, where the per lane performance of a SAS SSD is unmatched. It's the highest performance. You have some applications where you can connect an MME drive that has a wider interface, up to four ports, maybe eight ports in some ways, and those would be used in some very hot applications, but it comes at a cost.

MME high-performance SSDs may be more expensive. And you look at those trade-offs, and this is just to show that the flexibility of SAS allows you to service this hot tier to meet a certain price per performance metric. So what does 24 gig SAS offer as a benefit for this tier? Well, you support high performance SAS and SATA SSDs. If you don't want to go all out and get the highest performance SAS SSDs, you just need some performance that is not top notch. It's just to maintain a certain performance level. Maybe SATA SSDs are what you're focused on. So again, we're targeting the cost per performance point

here to suit the needs of our infrastructure. Then you have the warm tier. So this is a tier that's meant to be more scalable, more flexible, it'll be a combination of cascaded JBODs, multiple servers for redundancy or multipathing needs. You'll have some expanders implemented in these enclosures or even in the servers and providing some cascading or tree architectures to support your storage needs that are, you know, they're called warm tier because they're not really hot, but they're not really cold. It's more of a just right.

So it's a balance. You know, you allow the scalability to add more storage as needed, whether it's performance-based or capacity-based. But it gives you the balance of capacity, performance, and cost all in one architecture. So it's services of multiple needs, and it allows for redundancy as well. So the staff technology has a redundancy factor built in, where you can have multiple hosts connected to multiple parallel platforms. If you lose a connection somewhere, lose a server somewhere, you have another path to get to that data. And you want to maintain that so that you're always on in that manner. So the applications for this, the Internet of Things, where, you know, performance

may be important, but you're storing a lot of data too. So you can deploy this in IoT applications. Business analytics, you know, you need more capacity there. So you can store your data and then go in and analyze it. Whether you take that data and then take it to a hot tier to analyze it, there's different ways to approach that, but this is usually the tier it goes through to manage and process that data. And then, as I showed in the previous example, social media. This services a lot of needs for social media where you need some performance to access this stuff, but basically more managing the data

and deploying it through different tiers of the type of storage as needed. So kind of an interface to those different levels. So overall, how SAS benefits this tier is providing the dynamic scaling to support the near-line storage media up to 64,000 devices. Typically, in SaaS architecture, you don't see deployments more than about 2,000 devices, for example, to have the right balance of performance. But that scalability is there if all you care about is the access to that amount of capacity. You have that flexibility and that scale.

This tier also is benefited by SAS by allowing the support for hard drives to be introduced into the storage media. So you can have a combination of SSDs and hard drives to get the right balance. And ultimately, as I mentioned earlier in my discussion on this slide, liability where you have all the path failover and RAID applications deployed in this tier. And last but not least, we have this cold tier, scalable and flexible storage. And ultimately, what you're focused on here for your storage requirements are large scale.

You want to deploy a lot of capacity. You want it to be low power, low cost. And you want to have a certain dollar per gigabyte metric so that you're not spending way too much on your storage. You only spend as much as you need to deploy it. So the applications for this are social media, again, where you're relegating your older access, your aged photos and videos and posts to a server that is not accessed as often, but the users don't want to lose access to that data. They want to go down memory lane, for example, or the feeder server or analytics server can access that, and you guys may have seen some posts at the top of your feed

that says, remember this seven years ago. You're accessing that old storage data to remind you of what this seven years ago. You know, you're accessing that old storage data to remind you of what happened seven years ago, whether you want it or not, but that's some of those applications. And then cloud storage. I mean, you've got a lot of cloud storage that's used today with the COVID applications,

people working from home. You have more data that's being accessed now more than ever in your home environments, through the cloud, meetings, online platforms, virtual trade shows such as SDC here. These are being stored in the cloud storage area more and more often these days. And then IoT, again, you have some data you want to store, whether it's your ring doorbell video that's being stored in case you ever want to go back and look and see that bear

that came out the front door, for example. It's ever-increasing needs for storage, and SAS and SATA is there for it. So what does the 24-gig SAS offer to this tier? The flexibility and the support for SATA hard drives. It's still SAS in architecture, so you're still supporting the overall different types of media that are required for your specific application. So, again, that flexibility. And in your dynamic scaling, again, up to 64,000 devices, you may not get to that large quantity of devices deployed, but that is available to you if you deploy it in that way.

And then rack level and data center level reach. So with this type of deployment, supported by the SAS architecture, SAS specifications, you can do optical cabling, copper cabling, whether it's active or static type of copper cables. You have different connectivity options to deploy these in a large data center across the room. You can even implement SAM type architectures using different technologies. That's a front-end interface.

iSCSI private channel, for example, are some common front-end interfaces to deploy SANs that have SAS storage deployed behind them. So there's a lot of flexibility. So what does this all lead up to? Why talk about this now? So 24 gig SAS is the latest implementation of the SAS specification and it's coming available now. We're overall ready for the ecosystem this year. The ecosystem is on track for this production readiness, with some devices being available now, some coming towards the end of the year, depending on what you look at.

SAS 4 analyzers have been sampling since the last year or two. So that starts the ecosystem to do their development and testing to design their products. And then you have the cables and connectors that connect all this. Those are in a sampling stage to support those that are in their development cycles. And they support the new form factors that are designed for 24 gig sass and ultimately for 24 gig sass we're supporting mini sas hd sas mini link slim sas and got a couple of new form

factors there to meet the varying needs of the end users and those are all being sampled and built and productized now. And not to discount the other portion of the SaaS ecosystem that makes it interconnect and make it all happen, SaaS for controllers and expanders. You know, those are starting to, you're starting to see messaging about those coming to market. And ultimately, those are targeted towards Gen 4 platform watches, depending on your favorite ecosystem CPU platform vendor. There's different availability there, but for that transition into that next generation of speeds and feeds, the SAS 4 controllers are starting to align with those deployments and look for those in your future servers. And all of this comes together to support the new hard drive and SSD capabilities

that I mentioned earlier. Those are all intersected with the 24 gig SAS ecosystem, whether it's multi-link SSDs, hybrid SMR, multi-actuator, hammer-mammar technologies. A lot of those are background technologies, but these are all supported by a SaaS architecture, depending on the media deployment. So what to look for here is the Discussing Trade Association did host its first PlugFest. That was over the last year. We host its first PlugFest. Over the last year, we had our first PlugFest that was kind of an early adoption type of PlugFest to see 24 gig in action, but not necessarily a full-scale PlugFest. Just to kind of get the show off the road and get things kick-started. And now we're looking for a full-blown end-to-end 24-gig SAS type plug fest here in the coming year.

The initial target was for this year, but with COVID impacts and limitations to travel and getting people together, we're looking at our second 24-g gig SAS plugfest to be probably in the late Q1 2021 timeframe. Face to face if possible, or if we continue to see challenges in the current markets, we'll look at a virtual option

and support our user base as needed to make it happen. So we can move forward with 24 gig SAS deployment. So as some takeaways here for today, first off, I wanna thank everybody for watching this presentation and attending this event. But for some ideas to take away, just keep in mind with 24 gig SAS, you're addressing the accelerating demand for storage. I mentioned maybe a couple times, but

COVID is creating unprecedented demand for data, whether it's kids in school doing meetings, people doing work via meetings, doing virtual conferences such as SDC here, or streaming more. There's lots of data access that's happening today. And SAS is well- well positioned to support that particularly with with capacity requirements for cloud storage or for high-speed access as needed for streaming video conferences for example.

There's various ways that SAS is supporting this infrastructure and as you can see with them the market scale and the market data that I provided or showed earlier, SAS has not gone away anytime soon. It's here to support the users for as long as is needed and provide the most efficient cost and deployment model. So that leads to my next point here, which is SAS continues to innovate around the reliability of storage.

So what does that mean? I mean, there's new technologies that are continually being designed and updated in hard drives and SSDs. And the SCSI Trade Association and the standards organizations not sitting by and just letting everything fly by. We're constantly monitoring and maintaining and keeping up with these different technology deployments and seeing how we can implement support for those in SAS. So, you know, whether it's new hard drive technologies or SSD technologies, we will find a way to make something supported more efficiently or provide the applications, the hooks to make something work in SAS and SATA deployments. And then again, that leads to my next target point, which is SAS architecture is optimized for scalable, flexible deployments, service-wide tiers. So here, with the flexibility and the continuing adoption of new technologies, we're supporting different tiers that require different storage needs.

SAS is inherently scalable and flexible. You can deploy it in hot tiers that allow for the instant access that you require for processing on-demand AI needs or communication needs. Down to the warm tier, which is servicing your standard balance of capacity and performance across the architecture down to the cold cold tier where you're you're doing your cold storage your aging data storage thus access storage you know it's not in high demand but it's still required at some point in time so you got to store it out somewhere in there and it needs to be a large capacity database or bank, so to speak. And with all of these implementations, you have the built-in management, especially for the enclosures. So you're monitoring everything, making sure everything is staying up and running, staying accessible with your availability.

So it's a continuous support model for SAS for the current deployments in the world. And I can't reiterate this enough. SAS is ubiquitous. It's not going away anytime soon. It's got a good balance of price for the capacity. It's got a good balance for performance for most needs that are out there, especially in enterprise data center deployments. And even the channel, they still like to use data drives in a lot of their deployments today because it's a good cost point. It's just enough for their needs. NVMe is certainly a good technology to adopt,

and we see a lot of parallel cooperative deployments of NVMe and SAS happening in the future. Just as an example, there's U.3. They call it U.3. It's a kind of a universal connector specification, universal backplane management. It's another specification. These are going on in the SFF standards through SNEM. And these allow for cooperative interconnectivity of SAS and SATA and NVMe to allow that overall flexibility for those varying storage needs.

The vast deployment for enterprise and channel data centers is going to be SAS and SATA. Hyperscale might lean towards other technologies. It's based on the needs of the application. And 24-gig SAS components are available today, You know, it's based on the needs of the application. And 24-gig SAS components are available today, starting with sampling, starting to go to production. There's been several announcements over the last months and more to be announced going forward. So the Scuzzy Trade Association is looking forward to this next-generation deployment and supporting that model for the years to come

and continuing to look at where things can be enhanced and improved to make things more efficient and make the end users happy with their access to their data through our common customers. So in closing, once again, thank you. Thank you for attending this session. If you have questions or for more information about SCSI Trade Association or 24GIGS, that's for that matter, visit us at www.scsita.org.

And I'd like to thank you all for attending again today and I hope to hear from some of you. Thank you. Thanks for listening. If you have questions about the material presented in this podcast, be sure and join our developers mailing list by sending an email to developers-subscribe at snea.org. Here you can ask questions and discuss this topic further with your peers in the storage developer community. For additional information

about the Storage Developer Conference, visit www.storagedeveloper.org.