Object Storage a ‘Total Cop Out,’ Hammerspace CEO Says. ‘You All Got Duped’

By Alex Woodie

November 2, 2023

The mass adoption of object storage systems like Amazon S3 may appear to be a top achievement of the big data era, since we got essentially unlimited storage accessible through REST commands. But to Hammerspace CEO David Flynn, object storage is a “total cop out” that perpetuates data orchestration headaches. “That was the bargain you made with the devil,” Flynn said at the recent HPC + AI on Wall Street event.

The industry standardized on object storage because the market found it too difficult to do the right thing: turning standard NFS into a distributed file system capable of handling cloud scale, Flynn said September 27 during his keynote address at Tabor Communications’ 2023 HPC + AI on Wall Street conference in New York City.

“This is all about fixing the evil that file systems weren’t able to do cloud-scale,” Flynn said. “Let’s face it: the entire reason Amazon created S3–Super Simple Storage–was because file systems are so damn hard to make and even harder to make them scale that they dumbed it down and said ‘Let’s just use REST interfaces from user space and not try to get that high bar of having the OS know how to actually mount it and consume data from it.’

“Object storage was always a cop out, and you guys all got duped into thinking you have to rewrite all of your applications to use it just to get to cloud,” he continued. “And it’s because they didn’t do the hard work of making a real file system that OSes natively know how to use, and that’s still considered the conventional wisdom. ‘Cloud means rewrite everything to use object storage and talk to persistence from user space.’

Hammerspace CEO David Flynn at HPC + AI on Wall Street September 27, 2023

“That is a total cop out,” Flynn said. “It’s throwing the baby out with the bath water. But they had to do it. There was no other way to host the multiple tenants and the huge amount of scale that was needed in the cloud. But that was the bargain you made with the devil to get to the cloud as you rewrite that stuff.”

These were strong words, but with decades of experience in the enterprise IT and HPC markets, Flynn has learned a thing or two about high performance storage. Before co-founding flash storage startup Fusion-io, which was acquired by SanDisk for $1.1 billion 2014, Flynn helped build some of the world’s biggest supercomputer at Linux Networx. These travels showed Flynn a significant gap exists between the enterprise storage market and the HPC community, which he proposes to help fill with his data orchestration startup, Hammerspace.

AI’s Perfect Storm

The advent of AI has created a “perfect storm” of needs that will demonstrate that the object storage compromise is no longer sufficient, Flynn said.

Everybody now needs HPC capabilities to train AI models, but distributed file systems are the only way to efficiently manage I/O and keep GPU nodes saturated. The current status quo with object storage, which requires users to rewrite their applications from native NFS to use HTTP commands (i.e. REST) over the network, doesn’t cut it.

Data I/O remains a bottleneck to AI

While NFS is the defacto standard in the enterprise NAS business and is natively supported in Linux, NFS has had its share of false starts. The group behind NFS tried to address the statefulness issue with NFS 4.0, but ended up butchering the file system, Flynn said.

“NFS 4 took all of the sins of NFS 3–the statelessness and all of that–and tried to remedy it by adding statefulness so that you can cache stuff and get it more efficient,” he said. “But it was retrofitted in both the clients and the servers. And you ended up with all of the overheads and evils of statelessness combined with the overheads and evils of statefulness.”

For example, having a program create a file and then write it so another program on the same computer read it, took five round trips serial to the filer in NFS 3. With NFS, that jumped to 15 serial round trips, Flynn said.

“They kind of messed up when they introduced all the statefulness because it wasn’t well tuned to really be exploited by the clients and servers together, and it ended up just being massive overhead,” Flynn said. “We still pay the price today. You want to go to NFS 4 to maybe get better security or other things, and yet you end up paying a massive price in a performance perspective. Basically, NFS has from inception always kind of sucked.”

After he left Fusion-io, Flynn vowed to find a way to remedy this situation. He teamed up with Trond Myklebust, the maintainer and lead developer of the Linux kernel NFS client, and co-founded a new venture Hammerspace. Led by Myklebust, a Hammerspace co-founder and its CTO, the team took what was effectively an academic project to develop a parallel version of NFS led by Los Alamos National Lab and turned it into an enterprise-ready product.

“We introduced the NFS 4.2 spec. That came from my team here at Hammerspace,” Flynn said. “Thanks to Trond, we were able to trick it out to actually really work well, especially with our NFS 4.2 parallel NFS server.”

A Global Storage Abstraction

Developing a parallel version of NFS was important to Hammerspace, but it was but one step on its overall journey to the ultimate goal: creating a new storage abstraction that eliminates the limitation of physical storage and puts data back in the driver’s seat.

As a data orchestration layer, Hammerspace essentially connects any storage infrastructure, whether it speaks NFS, object storage, or even block storage, to an application. Whether it’s a NAS device from NetApp, Dell EMC, Qumulo, or Vast–whether it’s an Amazon S3 bucket or a Linux server configured as a storage node–Hammerspace can take all that data and make it appear to an application to be sitting on local storage, even if the data is sitting on the other side of the world.

Just as characters in Japanese cartoons can pull a mallet out of thin air and whack their opponents in the head, Hammerspace becomes the thin air (or the single global namespace) out of which you can pull any piece of data (just don’t hit your opponents in the head with it).

The key to this seeming magic trick is unified metadata, Flynn said.

“Hammerspace solves the seeming paradox of how can you have data everywhere and anywhere you need it, without ever having copied it?” Flynn said. “That’s not because it’s not local and accessed at local access performance. It’s because the metadata is unified. So while the data is physically distributed, the metadata is logically unified.”

Pulling the metadata out of the storage layer into a new layer that transcends those other layers addresses a litany of data management pain points, Flynn said. Data migration becomes a thing of the past when you can hook up Hammerspace to an existing network file system and instantly start accessing it through an application previously connected to a different storage repository, Flynn said.

Hammerspace implements a global data abstraction layer atop physical storage

“When was the last time you had to think about having your data when you move from one phone to the next, or when you go from your phone to your laptop or tablet?” he said. “Our consumer data already lives in a Hammerspace and the iOS platform, the Android platform have basically orchestrated your experience and your data for you. That’s what we’re talking about doing here, but for the petabytes to exabytes scale unstructured data that is behind everything.”

The headache of maintaining a high availability server configuration will be a thing of the past when Hammerspace is ensuring that data from one node can be accessed from another node. “You can get rid of all of these different forms of copy, like your data copy sync programs, the data migration tasks, when you’re going from one system to another,” Flynn said.

In the end, Flynn hopes that Hammerspace’s notion of data orchestration flips the script on modern ideas of data management. Instead letting data storage define what data is and what it means to us, the data orchestration layer defines the data once, and makes where it’s stored a mere implementation detail.

“Data orchestration differs from data management in a very simple way: Data management is what you do from the outside of the data presentation layer,” Flynn said. “The real evil is the fact that the file system is embedded in the storage system or service. The data presentation layer being inside of that storage system means that the data is really nothing but a mirage that is being rendered by that storage. And if you put it in different storage, it’s by definition different data, because its very existence is an artifact of the storage system or service presenting it.”

It seems very natural to say that your data exists in your NAS filer or your S3 bucket, because that’s what possesses the metadata and presents the data to you, Flynn said. But that’s actually getting it backward, because data, by definition, is a higher level abstraction than storage, which is just infrastructure. And therein lies the key thing that Hammerspace enables.

“Data doesn’t exist except for as rendered by storage,” Flynn said. “That seems very natural to say, but it is upside down. It means that the infrastructure is in charge and the platform layer is dependent upon it. But with Hammerspace, that changes because you pull all of the metadata out of the storage layer into something that can transcend any of those storage systems. So now you have where data can sit appropriately as the higher level thing that you focus on, and where the data actually lives, on which storage, can change.”

You can view Flynn’s entire HPC and AI on Wall Street presentation by registering at www.hpcaiwallstreet.com.

Related Items:

Hammerspace Raises $56M to Reimagine Data Orchestration

Three Ways to Connect the Dots in a Decentralized Big Data World

Hammerspace Hits the Market with Global Parallel File System

 

 

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

What We Know about Alice Recoque, Europe’s Second Exascale System

June 24, 2024

Europe officially announced its second exascale system, Alice Recoque, and you can expect to see that name on the Top500 supercomputer list in a few years. Alice Recoque is the new name for a supercomputer with the opera Read more…

Spelunking the HPC and AI GPU Software Stacks

June 21, 2024

As AI continues to reach into every domain of life, the question remains as to what kind of software these tools will run on. The choice in software stacks – or collections of software components that work together to Read more…

HPE and NVIDIA Join Forces and Plan Conquest of Enterprise AI Frontier

June 20, 2024

The HPE Discover 2024 conference is currently in full swing, and the keynote address from Hewlett-Packard Enterprise (HPE) CEO Antonio Neri on Tuesday, June 18, was an unforgettable event. Other than being the first busi Read more…

Slide Shows Samsung May be Developing a RISC-V CPU for In-memory AI Chip

June 19, 2024

Samsung may have unintentionally revealed its intent to develop a RISC-V CPU, which a presentation slide showed may be used in an AI chip. The company plans to release an AI accelerator with heavy in-memory processing, b Read more…

ASC24 Student Cluster Competition: Who Won and Why?

June 18, 2024

As is our tradition, we’re going to take a detailed look back at the recently concluded the ASC24 Student Cluster Competition (Asia Supercomputer Community) to see not only who won the various awards, but to figure out Read more…

Qubits 2024: D-Wave’s Steady March to Quantum Success

June 18, 2024

In his opening keynote at D-Wave’s annual Qubits 2024 user meeting, being held in Boston, yesterday and today, CEO Alan Baratz again made the compelling pitch that D-Wave’s brand of analog quantum computing (quantum Read more…

Spelunking the HPC and AI GPU Software Stacks

June 21, 2024

As AI continues to reach into every domain of life, the question remains as to what kind of software these tools will run on. The choice in software stacks – Read more…

HPE and NVIDIA Join Forces and Plan Conquest of Enterprise AI Frontier

June 20, 2024

The HPE Discover 2024 conference is currently in full swing, and the keynote address from Hewlett-Packard Enterprise (HPE) CEO Antonio Neri on Tuesday, June 18, Read more…

Slide Shows Samsung May be Developing a RISC-V CPU for In-memory AI Chip

June 19, 2024

Samsung may have unintentionally revealed its intent to develop a RISC-V CPU, which a presentation slide showed may be used in an AI chip. The company plans to Read more…

Qubits 2024: D-Wave’s Steady March to Quantum Success

June 18, 2024

In his opening keynote at D-Wave’s annual Qubits 2024 user meeting, being held in Boston, yesterday and today, CEO Alan Baratz again made the compelling pitch Read more…

Shutterstock_666139696

Argonne’s Rick Stevens on Energy, AI, and a New Kind of Science

June 17, 2024

The world is currently experiencing two of the largest societal upheavals since the beginning of the Industrial Revolution. One is the rapid improvement and imp Read more…

Under The Wire: Nearly HPC News (June 13, 2024)

June 13, 2024

As managing editor of the major global HPC news source, the term "news fire hose" is often mentioned. The analogy is quite correct. In any given week, there are Read more…

Labs Keep Supercomputers Alive for Ten Years as Vendors Pull Support Early

June 12, 2024

Laboratories are running supercomputers for much longer, beyond the typical lifespan, as vendors prematurely deprecate the hardware and stop providing support. Read more…

MLPerf Training 4.0 – Nvidia Still King; Power and LLM Fine Tuning Added

June 12, 2024

There are really two stories packaged in the most recent MLPerf  Training 4.0 results, released today. The first, of course, is the results. Nvidia (currently Read more…

Atos Outlines Plans to Get Acquired, and a Path Forward

May 21, 2024

Atos – via its subsidiary Eviden – is the second major supercomputer maker outside of HPE, while others have largely dropped out. The lack of integrators and Atos' financial turmoil have the HPC market worried. If Atos goes under, HPE will be the only major option for building large-scale systems. Read more…

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Everyone Except Nvidia Forms Ultra Accelerator Link (UALink) Consortium

May 30, 2024

Consider the GPU. An island of SIMD greatness that makes light work of matrix math. Originally designed to rapidly paint dots on a computer monitor, it was then Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Some Reasons Why Aurora Didn’t Take First Place in the Top500 List

May 15, 2024

The makers of the Aurora supercomputer, which is housed at the Argonne National Laboratory, gave some reasons why the system didn't make the top spot on the Top Read more…

Leading Solution Providers

Contributors

Nvidia Shipped 3.76 Million Data-center GPUs in 2023, According to Study

June 10, 2024

Nvidia had an explosive 2023 in data-center GPU shipments, which totaled roughly 3.76 million units, according to a study conducted by semiconductor analyst fir Read more…

Google Announces Sixth-generation AI Chip, a TPU Called Trillium

May 17, 2024

On Tuesday May 14th, Google announced its sixth-generation TPU (tensor processing unit) called Trillium.  The chip, essentially a TPU v6, is the company's l Read more…

Intel’s Next-gen Falcon Shores Coming Out in Late 2025 

April 30, 2024

It's a long wait for customers hanging on for Intel's next-generation GPU, Falcon Shores, which will be released in late 2025.  "Then we have a rich, a very Read more…

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

The NASA Black Hole Plunge

May 7, 2024

We have all thought about it. No one has done it, but now, thanks to HPC, we see what it looks like. Hold on to your feet because NASA has released videos of wh Read more…

AMD Clears Up Messy GPU Roadmap, Upgrades Chips Annually

June 3, 2024

In the world of AI, there's a desperate search for an alternative to Nvidia's GPUs, and AMD is stepping up to the plate. AMD detailed its updated GPU roadmap, w Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

How the Chip Industry is Helping a Battery Company

May 8, 2024

Chip companies, once seen as engineering pure plays, are now at the center of geopolitical intrigue. Chip manufacturing firms, especially TSMC and Intel, have b Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire