Access the podcast to hear from Molly Presley, SVP of Global Marketing at Hammerspace, Rob Renzoni, Sr. Director of Federal Sales at Hammerspace and Jon Friedenthal, AI Strategist at Cisco Federal, discuss how agencies can streamline AI deployments by integrating data orchestration, high-performance infrastructure and automation. Explore how Cisco Intersight's centralized automation and management, NVIDIA's AI technologies and Hammerspace's Tier 0 transform local GPU server storage into high-performance shared infrastructure for AI workloads.
[Anthony Jimenez]
Welcome back to Carahcast, the podcast from Carahsoft, the trusted government IT solutions provider. Subscribe to get the latest technology updates in the public sector. I'm Anthony Jimenez, your host from the Carahsoft production team.
On behalf of HammerSpace, we would like to welcome you to today's podcast, focused around how a unified data platform can accelerate HPC and AI initiatives by simplifying data access across on-premises and cloud environments. Experts from Carahsoft, HammerSpace, and Cisco Federal share how integrated infrastructure, AI software, and data management solutions can reduce deployment time, lower costs, strengthen security, and streamline AI operations. Molly Presley, SVP of Global Marketing at HammerSpace, Rob Renzoni, Senior Director of Federal Sales at HammerSpace, and John Fredenthal, AI Strategist at Cisco Federal, will discuss modernizing HPC and AI infrastructure with a unified data platform.
[Molly Presley]
So HammerSpace was designed originally with the concept that data has become incredibly decentralized. Our founder and CEO was the original technology founder of the company, Feud and IO, the NVMe card company. And as NVMe devices started to get put into local servers all over the place, data became more decentralized.
And he had this long-term vision towards how can we aggregate that, use all that available performance and capacity and data, and aggregate it into a unified kind of data plane was his original vision. We call it a data platform now. And you can imagine, we've been well over 10 years in development, and we've been in the market now for about five years with our solution.
And in that time, AI has come about in a very big way. And this technology for addressing high performance and decentralized data has become really needed in really any AI environment. So what we see is that the historical methods of using essentially copies of data to move data around between a data center and a cloud, or maybe two different data center environments, certainly introduces a bit of complexity around governance of keeping track of your data.
It introduces costs because now you need infrastructure to store a second, third, fourth, fifth copy of data. But then you might also be looking at, I want to use all this data in AI, and it's spread across different proprietary storage systems. So pick the vendor of choice of your file and object storage vendors.
They all have their own versions of data. And that data you want to be able to use in your AI environment. And it most likely isn't sitting right next to your GPUs.
If you're looking at maintaining a fixed budget, and your GPUs are of course going to be expensive, but maybe your SSD costs have gone up more than you expected, and even your hard drives have gone up more than expected, figuring out how to first manage your data itself and have as few extra copies of data as needed, but also know what data you have so you can get value out of it, are really the kinds of problems that Hammerspace is addressing with our technology.
What we essentially have done is first tackle the problem of proprietary versus standards-based. We do have our technology used in HPC environments around the world, national labs, U.S. government, as well as outside of the U.S. It's a great technology for that traditional HPC mod-sim type of workload where you need to have clients' software sitting on the client servers. You need a parallel file system that can manage keeping the metadata of the data path to deliver really extreme performance to your compute environment, whether it's an HPC cluster or a GPU AI type of environment.
However, the big difference that we've seen is historically those HPC environments were kind of bespoke, custom-built environments for whatever the compute cluster was being made for. You put in a network, you put in storage, you put in clients, you put in the compute itself, all for that fixed HPC environment, and it was designed to get really high performance across that entire environment. Now we're talking about this distributed data environment, so moving to standards has been really important.
Essentially our concept is that anything that runs on Linux should be able to be a member of our data platform and should be able to be a part of essentially the storage and capacity of our data platform. So you imagine that's our capacity is anything that's Linux, so Linux server, it could be your GPU nodes themselves, it could be other storage systems that are provided by companies like Cisco and our other partner companies. But using all standards where it's Linux and NFS-based or S3-based and doesn't require installing any extra software or using specialized networks in order to deliver the data to the AI environment is a fundamental piece of our solution as well.
We'll talk a little bit about how this looks. It's a little bit different than traditional HPC versus AI, and I've already talked about this a fair bit, but kind of conceptually. So instead of having this compute users and storage all sitting side by side next to each other in a data center, we now are bringing in and ingesting data from multiple sources and then likely moving the data once we've identified what the AI project is and what we want to do with it to the GPU cluster, which probably is getting data from five, six, a hundred different data sources.
The incoming data system is different, but once you get the data local to the GPUs, we have the exact same requirements that we had in traditional HPC of extremely high performance. You might have big files and small files in AI. We might call it something different.
Instead of it being scratch versus archive, we might be doing RAG versus training. But kind of the fundamental characteristics that we designed for as an industry for HPC are needed in AI. With a bit more complexity, we have a lot of metadata ops in most environments.
We have small files, big files. So we had to design our technology to be able to not only aggregate the data sets, but also keep those GPUs busy wherever we're feeding them. And I'll talk about a couple of customers as I tie up that talk about very centralized environments as well as very decentralized environments where they maybe don't even own their own compute and are using a NeoCloud or someone else's compute.
And I'll talk about that in a little bit. So that comes down to data orchestration, the idea of using data orchestration that once you know what data you have, you've aggregated into a single metadata environment in our data platform. Then you want to be able to really enrich that metadata, know which data do you have, have your tools that you're using for governance, able to apply that metadata as well so that you can say, here's my business rules.
Data of a certain type of attributes never can leave the data center. Data of another attribute can never be used to train a model. Those types of things are set as objectives within our data platform and essentially ensure that you can set the rules of how you want your data to be used.
And then we, through software and automation, ensure your data is only used the way you want to. So cool stuff as far as now you don't have IT people copying data sets from A to B, you aren't opening tickets, trying to get access to the data and figure out where your data is. I think you can understand readily how nice it is to be able to automate your data access, but being able to automate the tracking of what did get done with your data, where did it go?
Was it moved to GPUs or into a model? Those types of things are what we really look at from a governance perspective with our data orchestration of only placing the data in the environments, whether we're talking sovereign AI, government security requirements, we manage that as well. So you can imagine that being able to have in a single view of your data, knowing how it's been used, by whom and where it's going is really important.
And it's a huge step forward versus the idea of managing the data by copies, where instead of now saying, okay, well, I'm going to take the copy of my data, copy it out of the storage system where it was created somewhere else, all of a sudden you've lost track of where it goes. You don't know where your data went once it's copied out of that system. So there's a lot of really great things that we can offer in this environment by unifying the data to not just make your infrastructure faster, your HPC or AI jobs faster, but much better control of your data itself.
So what is Insider Data Platform? I'm just going to run through this for those of you who are more technical on the call and want to kind of know really what is Insider Data Platform. We are a global parallel file system.
So once your users, your applications, your models connect to Hammerspace, they have global view of all of the data and storage within our global parallel file system. So what that means is you can connect over S3, we have REST APIs, we have a parallel file system client, which is built into Linux. That's our PNFS client.
We have an MCP server. So you have many different options of connections to the data platform. What's really cool then is once that, once you're connected to our data platform, even if when we add a new storage system or a new cloud instance or a new Neo cloud where your GPUs are going to be, you don't have to reconnect your data and your users to those data sets.
They're already connected to Hammerspace. We just use rules of which data are they now allowed to see. So connect to us once, and then don't worry if you've migrated or moved data, you can see the data.
The data is then through a global metadata environment cataloged in a single metadata server. So you can see all the data, enrich that metadata so you can find the data you want. It's presented as a global namespace.
So we have global namespace that's an active read-write namespace that all those user models, given the appropriate permissions, can read and write to any of the data, regardless of where it's stored, if it's in an enterprise NAS, an object store, even tape or the cloud. And then the policies, we used service level objectives to have you designate how you want your data to behave, both from a governance perspective as well as from a performance perspective. You may have some really fast NVMe and some slower archive.
We can use those objectives to make sure your data is staying in the right tier or type of storage and in the right location. And then ultimately we can use the data that's sitting wherever it resides today. So for those of you who said procurement timelines and costs are an issue, if you want to make your data, and I kind of air quote this AI ready, without having to buy an entire new AI storage system, we can ingest, we call it data assimilation, the data and know which data you have in our metadata environment without actually putting the data into a new storage system.
So use the storage infrastructure you have, make that data AI ready. And then when you do buy new storage, you just simply add it to our environment. So you can work on making your data AI ready while you're waiting for your next storage system to arrive and take advantage of the investments you already have, as well as your cloud storage and other things that you have.
So that's kind of our stack from a high level. I do want to talk a little bit about our tier zero capabilities as well. So I've spoken a few times now about, we can present any type of storage in our data platform as this unified namespace under a single parallel global file system of capacity that includes the NVMe that's seen in your GPU nodes.
We call this tier zero. So we think of tier one storage as the type of storage you typically would buy and attach over a network to your GPU servers. It would still be quite fast.
That could be your NVMe storage systems, but local within your GPU nodes, you very likely have some NVMe that may not be used right now for anything extremely strategic in your AI workloads. And we can activate that and use that as part of our capacity pool as well. With the SSD shortage that's going on in the industry where procurement timelines are longer and costs have gone up a lot, we've seen this be a really great way for customers get moving faster on their AI initiatives while they're waiting on procurement.
And simply think about it. You've got your GPU nodes potentially already configured and available in your environment. It's essentially just activate the NVMe and you could be up and running in a day with this.
And so the idea of activate that fastest tier of storage that you have, lowest latency next to your GPU nodes, include it as your overall shared storage environment. And then all the software goodness that I've been talking about where Hammerspace will make sure that as data is written to those NVMe is if you need it protected to another node or if you need it tiered quickly to free it up to the next dataset coming in. There's a lot of things that we can do within our software data orchestration capabilities to be able to make sure that you adhere to your data protection policies, your data availability policies, and tiering requirements.
And so kind of as I tie this up, I want to talk a little bit about the customers that we have. As I mentioned, we do have both the Department of Defense, Department of Energy using Hammerspace, both in traditional HPC workloads, as well as in clusters that are being designed to be able to run HPC and AI simultaneously. So touching on the importance of being able to manage an AI environment in parallel with an HPC requirement, a lot of mixed use clusters are coming out now and our technology and performance attributes are designed very well to do that.
As we think about at scale operators of AI, Meta trained LLAMA2 and LLAMA3 and parts of LLAMA4 on Hammerspace. Some of that was done in their data centers with just GPUs and data and everything local to each other. Some was then using cloud-based GPUs as well.
So they took advantage of using Hammerspace in a hybrid as well as a cloud, as well as a data center environment. And then we have another large AI company, an enterprise AI video conferencing operator who uses us in a cloud-only environment. And they use us because of tier zero that they were already renting or using the GPUs in Oracle Cloud.
They could use the NVMe, the storage capacity along with those as part of the way objects in Oracle Cloud are priced. So they were able to essentially, for no additional money, use the storage that came with their GPUs and get this extremely high performance simultaneously. So it was very efficient and much faster than the parallel file systems that they had as alternatives in the cloud, so it's faster and more economical.
And then we have NeoClouds that leverage us, not just for high-performance storage, but even more importantly for the data orchestration to get your data into the NeoCloud. So being able to identify which data sets are going to the NeoCloud and make it local to the GPUs. And I want to just tie up with talking about the benchmarks that we released at supercomputing in the US supercomputing conference last year.
That we ran, you know, we talk about the importance of standards especially as we move into an AI world and the importance of making sure your infrastructure is fast. Lots of customers said, yep, nod their heads. You know, our National Lab customers and folks like that who have been tracking our technology for years now really agree with the need to move to PNFS and the standards-based, not just storage interfaces and the infrastructure, but standards in networking like Ethernet instead of requiring InfiniBand, things like that.
So we've begun to submit benchmarks, both the MLPerf, which we're not highlighting here, but the IO500, which we are highlighting here. This was the 10 node production challenge. And, you know, you might look at it and go, okay, you're excited about, you know, like a top 20 result.
Why is that? And when you look at who we're comparing against, this is a very long list of different environments that have run this very good apples to apples benchmark. And you can see we're saying right there, I'm above a bunch of Lustre deployments, below a couple of them.
And if you look at the larger list, what we really like to highlight is absolutely, if you look at the requirements that test all the different attributes of your performance, small, big file, metadata, et cetera, and in a very controlled benchmarking environment, you absolutely can achieve your HPC performance capabilities. You don't have to jump to Lustre or something like that just to get performance and then figure out how to also accomplish your AI strategy. We have the ability to deliver both even with decentralized data and even without any specialized bespoke environments on the infrastructure side.
And we'll continue to run these with other environments. This one was run with Samsung in the Samsung labs as one of our SSD partners who really wanted to highlight their SSD capabilities in these workloads. And with that, I believe I'm going to hand things off.
[Jon Friedenthal]
Thanks, Molly and team. Appreciate you inviting me here and looking forward to talking to you a little bit today. So several months ago, I was at a customer at the Department of War and I was with the CIO and CDO of this agency and NVIDIA was in the meeting with us and they basically unpromptedly gave this quote and I just love highlighting it.
It really brings to light like all the important things beyond a GPU. A lot of times our customers are throwing a lot of money at GPUs and expect it to work. And it's just not the way things operate.
There's a lot of components and a lot of things and that's going to be part of the theme of my conversation here is talking about all those different pieces and trying to get them to align and work together. And basically, as you can see here, their quote was, if you can't get data to your GPU, my product is a waste of your time and money. There is no company even close to providing the end-to-end secure, reliable data pipeline that Cisco came.
And basically, everything I'm going to share with you is Cisco earning that sentence or that remark. And so, candidly, idle GPUs are the most expensive line item in any AI program and agencies are not GPU constrained anymore. They're data pipeline constrained and knowledge constrained on how to deliver capabilities.
And so, in this conversation, there's three of us I'm referring to. So, what does each bring to bear? And at Cisco, we're going to be the pipeline, the perimeter and the compute.
NVIDIA brings the AI software and HammerSpace is the data layer that makes the other two worth buying. And Cisco delivers that end-to-end platform for AI. HammerSpace is the piece that can deliver the AI data pipeline so you have end-to-end secure, reliable performance.
So, as you can see here, whether it was by happenstance or design, Cisco invested in the last several decades in four key pillars. There's a few other technology areas as well, in collaboration and things like that. But we've been heavily focused for 45 years in networking, security, for multiple decades, compute since 2009 and observability via our acquisition of Splunk since 2003.
If you think of that in today's terms with AI and HPC, you need all these things to work together. And if you have any one of them faulting or not working properly when trying to deliver HPC or AI workloads, you're going to have problems. The ideal scenario is to have this environment working together as one and optimized as one.
And so, one thing you'll note is storage is an on the system. And so Cisco, we bring a software-defined data layer instead of a proprietary appliance and that's where HammerSpace comes in. We'll talk a little bit about today's data challenges.
So as Molly talked about, you've got disparate data everywhere and currently 90% of data out there is unstructured and not AI ready. And so in an agency, it's fragmented across mission systems, classification domains and geographically separated. And then petabyte scale data is like too expensive to me.
And AI behaves poorly with wrong or stale data. So this data pipeline or this data platform is absolutely critical. And the mission has problems in that they simply can't get the right data to the model without copying it three times over as Molly alluded to.
And when you do that, you have to re-accredit or re-ATO some of this stuff. And it's just messy, expensive. It's not scalable, it's not viable.
Here is kind of the core functions of a storage system. Again, this is re-going over some of what Molly already said, but you have your enterprise data organization, the core data management governments, the preparing data for AI, AI data management services and advanced caching and security. I'm going to call out a couple of things.
So specifically the KV caching or the tier zero with emergence. So in essence, instead of re-computing context every inference call, keep it on local NVMe inside the GPU. That's throughput you already paid for and aren't using.
AI data management services, meta-tagging, classification, masking, redaction, sensitivity analysis. This is how you make data AI ready without creating a new copy that needs its own ATO. So what's our role here at Cisco?
This whole platform runs on Cisco UCS. It's managed by Intersight. It's fed by Nexus.
Essentially Amberspace is the software. Cisco is a system that runs on it and Fabric it runs across. So in HPC environments, one thing I've discovered in talking with many customers is oftentimes these systems take years, literal years to deploy an ATO for their environments.
And what we're trying to do is solve by making this simpler and get the time to production, the time to value, the time to first token, all that stuff a lot quicker. And so we're going to talk about some of the things that Cisco is doing to make this 90 days to production. So if you build out a system and you build it out manually, there's all kinds of components.
We've touched on some. There's the compute, there's a software layer, there's the OS layer, there's Kubernetes, there's storage, there's all this stuff. And then beyond that, there's all these different things that you have to consider.
It just becomes really messy and it becomes very difficult and trying to get all the firmware stable and all that stuff so that they all operate together. This takes forever and it's really difficult to manage. So basically what we've done is we've built three, there's three consumption models out there, right?
You can build your own. There's AI pod, which I'll describe a little bit more with on-prem network management. And then there's the AI pod with cloud-based network management.
Typically we see our government customers choosing the middle one for IL5, IL6 and disconnected environments. This is why we developed the secure AI factory with NVIDIA. It's a modular reference design that takes the guesswork out of AI infrastructure.
Instead of spending months trying to stitch together disparate tools, we provide a pre-validated stack that integrates all of our high-performance infrastructure with full stack security and observability with NVIDIA's AI software and accelerated compute. So as you can see, amongst the stack, you've got at the top layer, you've got the AI software to the Kubernetes platform to networking compute and data storage. So at Cisco, we don't force a Kubernetes like distro or model.
So like if you're an agency that's standardized on OpenShift, for example, keep OpenShift, we're fine with that. The two columns on the right are some of our key differentiators. So it's the AI security and observability, which they run vertically across every layer and it's integrated throughout.
Every other vendor in the space bolts security on afterwards. And with AI, it's kind of created a threat surface that didn't exist before. So the secure is not just marketing, it's actually architecture.
So we embedded security and observability into every single layer from the silicon and the network to the AI models themselves. And with Splunk, you get a single dashboard to monitor the health costs and security of your AI tokens in real time. We help you build basically the true AI factory that doesn't just produce AI, but produces trusted AI.
One other big thing here is we're one accountable vendor. So when a training job fails at 3 a.m., there's one number to call and not seven pointing at each other. It's a real risk reduction and a value that's unique to us.
So how do we make this valuable specifically to government, right? And so with NVIDIA, we jointly have produced and are deploying the secure AI factory for government. And what does that mean, right?
It means you can put your own posture and that can be just a STIGs or your own agency's posture or preferences within a model. And within this catalog of models, basically we can enforce and make sure that not only that these postures are applicable and visible and working, but that the model itself doesn't get broken when you push STIGs down, right? Which is a big problem.
So oftentimes DISA has all these different rules, all these security postures and all that, and we actually implement them. You can get approval, but it doesn't work. This guarantees not only that it's meeting the security, but it's meeting the actual function.
So switching gears a little bit, let's talk a little bit about network efficiency. So there's four costs and I'll highlight a few of them, but the GPU idle time. So a GPU waiting on data, it still draws power.
It still burns depreciation and it still occupies rack and cooling dollars, right? And so you have that. And then AI job delays.
So network inefficiency, it doesn't show up as a network line item. It shows up like, hey, we need more GPUs. And so infrastructure teams invest more in GPUs when that's not actually the problem.
And then looking at like all the way to the right here, the energy tax, right? Power and cooling are frequently the actual constraint on how much AI can host. So all of these can be avoided with a well-designed and well-implemented network.
So to address that, Cisco NVIDIA came together and we brought three commitments. One is Spectrum X to include all Cisco Silicon and Cisco Optics, right? We have built Nexus switches powered by Cisco software and NVIDIA Spectrum.
And we are delivering the highest performance ethernet for AI with tighter integration. So what this is, is it's not a reseller agreement. We actually co-engineered our solution.
So two of the best companies positions to compete on Silicon chose to combine instead of compete. And so we're combining our solutions, co-engineering them and bringing immense value to customers. Cisco Nexus is powered by NVIDIA Spectrum X as we just talked about.
It is the only third-party switching platform that enables NCP reference architecture compliance. Also, most network teams already know how to run Nexus. They know how to manage Nexus.
AI networking does not have to mean a new team, a new tool and a new set of complexity or new interfaces. In summary with NVIDIA, we're bringing a few things. So it's AI scale performance, advanced virtualization and simulation, actionable visibility and telemetry via open APIs and end-to-end synergy with Bluefield DPUs and ConnectX SmartPix.
Stack automation is basically software that was developed by our partner, Quali. It's exclusively used for us on Cisco infrastructure and it's at no cost, no additional cost to customers. But it's software to automatically deploy, configure and manage all the layers of an IT environment.
So from the underlying hardware and network to the operating system, database and application as one single cohesive unit. So instead of manually setting up a server, installing software and configuring settings one by one, use code to define the entire stack at once. And so this is really cool if you have certain workloads, you can say, hey, I want to deploy a new Hammerspace node and I want it to be X petabytes, I want it to do...
You put in some basic configurations in here, it's going to tell you how much it's going to cost, what resources it's using, and then it will automate the deployment down to the network settings and everything in a matter of minutes to hours instead of days and weeks. And so you can deploy these complex environments in just minutes. So this brings three important things.
It's speed, consistency and then scalability. It makes it easy to spin up or tear down resources as your demands change. One thing of note in the next slide here is there's an entire ecosystem of compatible solutions, right?
But if you have your own solutions that are built in house, those can also be leveraged through this automation platform, stack automation as well. So if there's existing workloads that might be custom made or aren't part of this list, they can be added to it as well. Whatever's already in your environment, Red Hat, NVIDIA, HammerSpace, the ISV layer that was above there, that's validated ecosystem, but it's not a closed.
So to summarize on three simple things, right? Trusted is three things. It's simple.
So Cisco, NVIDIA and HammerSpace already did the integration. We have CVDs, which de-risk deployment. The ordering is simplified.
The deployment is automated. It's scalable. That means high performance enterprise AI infrastructure.
So it's pre-validated, modular, compliant, and it's end-to-end from the silicon all the way to the Kubernetes, and it's secure distributed security fused into every layer, supply chain to runtime, models to agents, to infrastructure. And basically your AI team should never have to think about the infrastructure, right? The mission teams should be focused on the mission and the infrastructure owner gets to be the one that makes that possible.
With our joint solution, the data gets to the GP. So HammerSpace is parallel global file system and tier zero architecture. We put the hot data on local NBME inside Cisco UCS GPU nodes.
Cisco Nexus provides the low latency load balanced scalable fabric underneath. It's one system, not three products. So Intersight manages the UCS state and the HammerSpace cluster.
It's one lifecycle, one API surface, cloud-like operations on-premises, and then it's validated and defensible. It's a repeatable blueprint that eliminates the data silos, the security and observability is designed in, and it's a single accountable support path through us. NVIDIA makes the GPUs fast, HammerSpace makes the data findable and close, and Cisco makes it reachable, secure, deployable, and operable, and Cisco is the only one that can do all four.
[Rob Renzoni]
You made AI sound so easy just now. Coming from a traditional HPC architecture landscape, HPC has always been so complex, and to your point, took years to deploy a new HPC cluster and then all the bits and bytes that go around that, getting it up and running and certified. What we've had the opportunity recently over here at HammerSpace to do is work with our DoD mission partner to help them bridge that gap and move from traditional complex legacy HPC clusters over to these simple AI pod architectures, or as we're calling it for DoD, an AI factory, that takes not only their traditional HPC workloads, but also the new emerging AI tools that are being leveraged out there in the field to give them a full stack of options as far as their HPC clusters go. With Cisco, NVIDIA, and HammerSpace, we were able to design this architecture here that I'm showing you.
It's a powerful stack, and it really enables the data scientists to leverage the latest tools from run AI to seamlessly utilize the data orchestration capabilities that Molly had talked about, along with driving extreme GPU efficiencies that you want to have in these kind of environments because idle GPUs cost money. The most expensive piece of this whole architecture is those GPUs, and as we worked with our mission partner here, we helped them optimize their ability to purchase more GPUs because that is the hardest commodity to get and the most expensive and pretty much the key to the whole puzzle by implementing a tiered architecture, leveraging not only that tier zero capability, but also flash and traditional hard drive architecture to give them everything that they needed to, but lower the cost of the storage aspects while allowing them to purchase more GPUs.
And you can see in this use case, not only are we using flash and HDD, but we're also mixing up the GPU types because Cisco is able to deliver different types of GPUs. This aligns to the customer's need to have different tools being run at different times and different workflows and divide up and conquer because this is a resource that's going to be shared across multiple entities within the DoD community. So overall, I think this is the way to go.
It makes a complex situation really easy. You have new folks entering into the AI market, looking at AI infrastructure, trying to figure out what to do. The Cisco AI pods are easy and simple to deploy.
And then for those legacy customers that are in traditional HPC, the tools that Hammerspace has around assimilation and easing into a new architecture make it a no brainer from implementing a new parallel NFS file system.
[Anthony Jimenez]
Thanks for listening. And thanks to our guests, Molly Presley, SVP of Global Marketing at Hammerspace, Rob Renzoni, Senior Director of Federal Sales at Hammerspace, and John Fredenthal, AI Strategist at Cisco Federal. Don't forget to like, comment, and subscribe to Carahcast and be sure to listen to our other discussions.
If you'd like more information on how Hammerspace can assist your organization, please visit www.Carahsoft.com or email us at hammerspace@Carahsoft.com. Thanks again for listening and have a great day.