5 Steps to a Smarter Private Cloud, APJ | Flavio Fomen, Intel – Sponsor perspective

Clips
More from 5 Steps to a Smarter Private Cloud, APJ

Flavio Fomin

Consultant Engineer on Storage and HCI

Intel

Rob Strechay

Dir./Principal Analyst & Host

theCUBE Research

play_circle_outline Transitioning from Hyperconverged Infrastructure to Disaggregated Private Cloud: Enhancing Operational Efficiency Beyond Current Limitations

play_circle_outline Unlocking Efficiency: Dell Private Cloud's Automation and Innovation in Resource Management

play_circle_outline Software licensing issues arise with appliance models of hyperconverged infrastructure.

play_circle_outline Flexibility in hardware and software licensing provides organizations with multi-hypervisor capabilities.

play_circle_outline Quick Assist technology offloads CPU-intensive functions, increasing efficiency for storage and network tasks.

Info
Transcript

Flavio Fomen, Intel – Sponsor perspective

Flavio Fomin

Consultant Engineer on Storage and HCI Intel

Rob Strechay

Dir./Principal Analyst & Host theCUBE Research

HOST

Dell Technologies “5 Steps to a Smarter Private Cloud” coverage continues, as theCUBE’s Rob Strechay is joined by Flavio Fomin of Intel Corp. to examine why enterprises are moving beyond hyperconverged infrastructure into Dell Private Cloud. Fomin explains how a composable, disaggregated model preserves HCI-style automation while unlocking stranded storage and decoupling software licenses from hardware. He also highlights rising demand for multi-hypervisor strategies and a bring-your-own cloud operating system approach that curbs lock-in and aligns cost contr... Read more

explore Keep Exploring

What is the significance of hyperconverged infrastructure and how does it relate to the emergence of DPC as an architectural innovation? add

What does DPC stand for and what are its key benefits in relation to resource management and cost performance? add

What are the challenges associated with the appliance model in hyperconverged infrastructure regarding software and hardware licenses? add

What advantages does the new private cloud model with Dell provide to organizations regarding their cloud operating systems and workloads? add

What is Quick Assist technology and how is it utilized in various industries? add

bolt Powered by CUBE AI

Flavio Fomen, Intel – Sponsor perspective

search

Rob Strechay

>> Hello, and welcome to this segment of Dell Technologies' Five Steps to a Smarter Private Cloud. I'm joined by Flavio Fomin from Intel, who's also... Intel is one of the sponsors of this entire series. So welcome on board, Flavio.

Flavio Fomin

>> Thank you so much, Rob. Thank you.

Rob Strechay

>> So I think one of the things that we find very interesting is how private clouds are really transitioning and there's a modernization going on with them. And there are differences between customers and partners who may have looked at hyperconverged infrastructure, or HCI, in the past, and now we're really looking at things like disaggregated infrastructure. What are you hearing, and what's Intel's view on why this move makes a lot of sense for customers as they evaluate the transition with their private cloud?

Flavio Fomin

>> All right. Yeah, this is a great question, and I think that you start to address that you come up with an answer here. It's important for us to first acknowledge the value that hyperconverged provided to everybody over a decade or a little bit over a decade, that significant improvement in operational efficiency that we get from hyperconverged. And then if I look at this short history, we saw a lot of incremental value, a lot being added in every new release of the hyperconverged. But like any technology, what we're going to see here is that at some point, that incremental innovation stops addressing some of the main pain points that the organizations have. And that's how I see DPC. I see DPC, being as a architectural innovation, as a bigger jump to address some of those pain points.

Rob Strechay

>> So before we go further, what does DPC stand for, just so... People might not know that acronym.

Flavio Fomin

>> DPC stands for the new Dell Private Cloud. We have been looking at this from Dell's perspectives for a while. And it is a new technology that gives you that higher level of automation, that experience that you have on hyperconverged, but opening up some of the innovation opportunities. But thinking on those pain points, I think I'm very strong about two of them. One is what I see as these trapped resources, trapped storage resources. The other one is this whole of flexibility of hardware and software licenses. In this scenario where I have two clusters, at one cluster that is starving for resources, for storage resources, and while the other one has excess capacity over there. But that excess capacity is trapped I cannot share across that. And how you solve that? Well, one of the ways is let's use an external dedicated storage system. But I want to do that without losing all those benefits that I got on hyperconverged. And that's one of the things that I get from DPC, from this new automated platform that is being introduced. And by doing that, thinking in value, I get very tangible cost performance benefits out of this transition type. So that's one. There is a debate. There is a discussion today about software licensing in this ecosystem. And when you talk about that, I think we have to put in perspective that hyperconverged can be deployed in more than one way. But when I think in the appliance model, which is really the prevalent model, in this model, it is a typical approach for everyone to tie the software license to the hardware license. And this is something that has a purpose, had a purpose, that you want to make sure that that final product has a super high-quality, is going to work as expected. But as I was saying, that this became a pain point because over time, what organizations are saying is, "Hey, I want to upgrade my hardware and take my software license to the new one in this tech refresh." It's a pain point because I can't do that. Or similarly, "I want to get this hardware here and I want to reuse that, repurpose that, and put that as a different hypervisor, and I can't do that." This is the other thing that this new private cloud model that we have with Dell is addressing. In this case, the business model that we use allows you to bring your own cloud operating system and gives that flexibility. Now, why is this important? It is mostly because, what I hear from several analysts, is that organizations... For surveys. Number is like 90% of the organizations, they are saying that multi-hypervisor is or has become a strategic imperative for that. So this is defined.

Rob Strechay

>> Yeah, we've even seen that in our research as well. So, yeah, I think to that point, it definitely is... People are looking for choice, and they're also looking for the best to help them with the different workloads that they have, and I totally agree. And part of it is that disaggregated nature because of the storage being locked into one place.

Flavio Fomin

>> And that's really the value that I see, Rob. I kind of prefer to focus on the composability portion. So I have all that pool of resources and I can draw resources from that pool, create my private cloud, and well, if I need to implement some elasticity concept here, I don't need that research anymore, I return that to the pool, and I can reuse that with another hypervisor. So on the value side, it's really... As I said, there are a lot of different areas of value, but this cost performance that I get, this additional cost performance from the storage is really powerful because it extends across all of those different hypervisors, this new concept. Second is this overall strategic value of flexibility that everybody is looking for. And I see, that is the thing that requires these architectural innovations that we are seeing. Dell is innovating in the automated side of that in a really big time, giving everybody the opportunity to enter in a new wave of innovation.

Rob Strechay

>> Yeah. I totally agree with that, and I think that the movement from HCI, it really led to what we see as the platform engineering and the DevOps revolutions, SREs, and a whole new way of really operating like cloud and bringing that on premise and being simple. I think there's also some stuff... You had some recent Intel innovations that have really helped to benefit people who are moving to Dell Private Cloud. Kind of unravel that a little bit for us as well.

Flavio Fomin

>> When you look at CPU recently, one of the core performance indicator, the key KPIs, is performance per watt. And if I look at... Let's use PowerEdge here as our reference point, and PowerEdge 14G to 15G to 16G. So in the transition between these generations of PowerEdge, we saw an improvement of about 20% in performance per watt. Now, with PowerEdge 17G, which is where we have our new Intel Xeon 6, we are seeing a much bigger jump. It's more than 40% improvement in performance per watt. Now, if I go a little further, I look into minus two generation, the 15G to 17G, which is kind of where people would be thinking of when they're thinking on tech refreshes, we're talking of near 70% improvement in performance per watt. So to put this a little bit in a bigger perspective, I'm kind of a performance engineer at heart, so I like the numbers. If we take the PowerEdge 15G, the top being CPU there versus the top being the 17G, we are seeing here a more than three times increase in number of cores; we are seeing it's about four times increasing the overall CPU capacity, with near 70% improvement in performance per watt. So why do we care about that? Let's say, back to the DPC point, if I'm considering that transition, there is a tremendous opportunity for server consolidation in there, and that adds a lot of value when I'm thinking on the constraints that every data center has today of constraints in power that they have rack and in power in the overall data center. So you have that opportunity now to improve in a significant way your overall CPU capacity without... and improving also the overall power consumption.

Rob Strechay

>> And that makes complete sense to me when you look at consolidation and you're looking at how organizations are bringing these together in a private cloud way. What happens when they look at it and say, "Hey, I really need a dedicated server"? So the performance per watt per server is clear when you're talking about a consolidated server scenario, but what happens when it's, "Hey, I have a workload that I need a dedicated server for"?

Flavio Fomin

>> So it is a good point because my previous comment, answer, really focus on that general purpose VM implementation. I have a VM farm, and in my typical tech refresh, I can implement a server consolidation strategy to optimize. But you are right, there are a good number of workloads that they still require dedicated servers, be that because they need to isolate that workload or because they need dedicated resources for that workload. So for that scenario, Rich 1 Socket solution, and it's Rich 1 Socket because it has 50% more PCIe lanes than the standard single-socket alternative. So what that means is I can have now a single socket that can have a lot more PCIe cards while still using or configuring up to 32 NVMe storage drives. So this is something that creates opportunity for several workloads, in thinking really real-life workloads that have an opportunity to increase their overall capacity, while at the same time reducing the total amount of power that the CPU is consuming, in part because of these performance per watt optimization that I mentioned, and the other part is really because it's going from two sockets to a single socket. So that would be a nice innovation here for that use case.

Rob Strechay

>> That makes a lot of sense as well, following the power curve, using less and consolidating there and being able to do it in one core. That really makes a lot of sense for these types of dedicated infrastructure. But beyond core count and core speed, what's Intel got going on?

Flavio Fomin

>> Got lots and lots of features that we have in the CPU, right? We have to look at the CPU as a system. For the sake of time, I have to pick a couple of them. So let me say one is FIDO; it is about that new secure and also hands-free, a zero-touch onboarding of devices that is implemented into the DPC. Intel provides full support. Intel was a pioneer in the development and provided the technology as a standard really to the industry continues to support that fully in the Xeon 6, so-

Rob Strechay

>> And it ties back into the whole confidential computing aspect of everything.

Flavio Fomin

>> It enables that element of security in the deployment. But then extending that to the confidential computing, we have a feature that is called TDX. So TDX stands for Trust Domain Extensions. So when I talk about that, it's really about having the hardware isolating the virtual machine from the hypervisor. And then you have also the Intel total memory encryption feature that is used to encrypt the memory that is in use. So putting that in a bigger perspective, we have been around for a while, so we have seen the whole investment that the industry put on data at-rest encryption, and then we saw data in-transit encryption. And now what we're seeing is really the data in-use encryption under this confidential computing scenario. So when I think of that, I think of use cases here, a very common popular use case in there is a multi-party analytic report, and I think of those highly regulated industry, I think of finance and healthcare. The analytical report contains data that multiple people inside of the organization can see, but the raw data is confidential, right? Only specific department there can see in the other one, and they cannot share it with each other. So in this context of confidential compute at those departments, they send their confidential, encrypted data to a shared platform, a virtual machine that is composed as a trusted execution environment, and then only the applications running those VMs, they can access the confidential data to produce that report. This is one of the elements that you implement with TDX. Now, long story coming here to the key point, this technology already exists for a few years, but the new news is that many hypervisors, they are adding support to TDX right now. This year they already have what they call a tech preview support, and then in upcoming releases soon, they will have what they call full support, and the difference really being that full support will enable live migration of those VMs. And this is something that is making confidential computing a much easier thing to be deployed in environments like DPC. If I have time, I have one more.

Rob Strechay

>> Go right ahead.

Flavio Fomin

>> Quick assist is really a technology that we have for offloading specific functions that consume a lot of CPU, so two of them; one is compression, and the other is encryption. So this is a technology that we have that in PCIe cards, and what Intel did was get that functionality and move that inside of the CPU. And this has been used by storage products, by cloud service providers for storage compression, has been used by telecommunication, telco companies, network companies producing... that are offloading encryption in there. And like in the TDX case, the hypervisors are now providing support to this technology, which enables us to see and to take advantage of that in our private clouds. Now, what is the point there? The number of use cases right now is limited because we don't have yet that full hyper... We haven't had, and now we have it. But I think the other element is it's a lot of value. When I think of these accelerators that are there that can allow you to reduce a lot that CPU utilization, and you're already paying for the CPU... And this thing is already there. You don't need to buy that. I can come up with two use cases very quickly here. One use case for compression, for example, everybody is running AI inferencing today, and everybody that is running inferencing is very likely using a KV cache because you don't want to be recomputing every new token. Using your cache, your inferencing time is four or five times faster. Now, problem is that KV store keeps getting bigger and bigger as your language models are also getting bigger. So everybody's deploying at different types of techniques to compress that, be that quantization, sparsity, pruning. But they also use software-based compression like Zstandard to compress that store. That compression is consuming a ton of CPU. So if you can offload that activity to this accelerator, it's a pure value that you're getting. And again, it's there in the CPU. So a couple more, but I'll...

Rob Strechay

>> Yeah, no, I think that that's a great place to leave it. I think everybody is dealing with AI and the sprawl of data and compression and how that really pulls together. But I really thank you for coming on board, Flavio. This has been awesome. We really learned a lot here, so thank you.

Flavio Fomin

>> My pleasure. My pleasure. Thank you.

Rob Strechay

>> Stay tuned to Dell Technologies' Five Steps to a Smarter Private Cloud. We'll be back here on theCUBE, the leader in analysis and news.