Perspectives on Grid: Improving Data Access Performance

By By Matt Haynos, Program Director, Grid Technology and Strategy, IBM

March 13, 2006

** First published by IBM developerWorks at http://www.ibm.com/developerWorks. All rights retained by IBM and the author.

A lot of what grid is about inevitably involves data, information and the intelligent processing of both across a distributed infrastructure. But often, data takes a back seat to grids focused on computation, at least from a publicity perspective.

Why this is? Often, those familiar with grid computing cite popular grid initiatives, such as Berkeley's SETI@Home or IBM's World Community Grid. When the World Community Grid was launched near the end of 2004, ComputerWorld wrote that IBM and leading healthcare officials had “launched a global grid computing project aimed at harnessing unused global computing power to help solve a variety of health issues and other scientific problems.”

But it's the data that matters. If and when intelligent life is uncovered by SETI, few people will focus on the millions of computational resources that were used: “Wow, we discovered life on Planet X, but, more importantly, we used 13 trillion MIPS from across the planet Earth to discover it!”

In this article, I'll highlight technologies, approaches and overall trends being used in the industry to intelligent manage data in distributed grid environments. There are a variety of technologies and approaches organizations can employ; the number of methods can lead to confusion when you try to determine which ones are best to use in which situations and environments. I'll present a survey of some technologies and approaches grid users employ when tackling the integration of data with computation and dynamic resource management, and I'll uncover some interesting trends in the marketplace. This article is not a comprehensive treatment of this subject, but rather a broad overview, designed to provide a firm knowledge foundation for grid-based data management.

In many respects, intelligently managing data access, movement and transformation is the key to successful grid deployments. This is especially true as grids grow, new applications and users are added, and the scale increases. Data becomes more and more of a concern. In some industries, such as electronics, the sheer volume of data is so large that data architecture, flow and network design are the primary grid design considerations.

So, you think you have data-sharing problems?

There's been a lot written about the LHC (the large hadron collider coming online in 2007 at CERN in Switzerland) and the associated applications (really experiments) and the LHC computing grid. But the key point is that the LHC computing grid was built, first and foremost, from a data-sharing and data-access perspective. Everything revolves around getting an unfathomable amount of data from the collider out onto the grid and to scientists worldwide.

The architecture of the grid uses a layered approach to stage data across multiple tiers of sites across the world. And it's a good thing it does, considering the scale:

  • Currently, there is something on the order of 13,000 processors on the grid.
  • In 2007, as the grid comes online, it is expected to grow to 130,000 processors (and very likely beyond that number).
  • The collider will generate 1.5GB of data per second for 10 years.

In case you weren't counting, by my estimates, that's 675 million CDs or 807 billion novels — enough books to stretch 13 round trips between the Earth and the Moon!

Emphasizing the LHC here is critical because this system highlights the importance of architecture, and why it is imperative to think through and understand data issues at the very earliest stages of grid design.

The technologies involved

There are myriad core data technologies that can be used in distributed grid environments. By themselves, they don't solve data-access and data-sharing problems in grids, but a well thought-out architecture does that. Effectively leveraging data technologies at the right place in the architecture is essential to streamlined grid operations. These data technologies are sort of like tools; you have to know when to pull the right one out of the toolbox.

Data technologies applicable to grid environments consist of:

  • Replication.
  • Caching.
  • Efficient data movement.
  • Data transformation.
  • Global naming.

Let's look at a few in a bit more detail.

Distributed file systems

While I didn't explicitly call out distributed file systems as an applicable technology, they offer many of the core technologies in an integrated package. Using distributed file systems as a core technology for distributed computing and grids makes a lot of sense because users needn't worry about implementing and integrating a lot of the subtechnologies.

It's really convenient when a distributed file system hides the underlying technologies from users. They provide a level of abstraction and convenience that is appealing in complicated environments like distributed grids with many applications and data sources.

Many sophisticated grid environments are built on distributed file system backbones. For example, IBM's systems and technology development groups rely on a combination of General Parallel File System (GPFS) and Network File System (NFS), as well as Andrew File System (AFS), to support some really complicated microprocessor and system designs with large data requirements. The underlying file system technology is utilized to support caching of data over remote networks, as well as to provide a common way of naming or referring to files. The TeraGrid, a grid connecting the U.S. supercomputing centers, employs a variety of file systems (including GPFS-WAN for parallel file access) to support experiments and simulations.

There are a lot of file systems out there. Some are meant for clustering (to share information among a cluster of nodes), others are designed as full distributed file systems with lots of advanced features.

NFS, though, has always been a “standard” distributed file system, crafted in some ways with the purpose to level the playing field across many physical file systems. This is an important consideration in larger enterprises in which many types of physical file systems (those that run on a machine or set of machines) are likely to be used. With Version 4 of the venerable protocol, provisions were made to support more sophisticated data performance and security technologies. This was a problem in the past with previous versions, but with Version 4, support for a feature called client delegation allows for implementers to provide caching functionality. As NFS V4 continues to mature, it will be a key technology in the distributed filesystem space for grids.

Distributed memory caching

Distributed memory, or application caching, is an interesting technology that has gained in popularity with grids. Some people even refer to application caching as a grid. Products in this category can significantly improve the throughput to underlying storage and reduce overall database transaction bottlenecks.

Products such as IBM WebSphere ObjectGrid, GigaSpaces' Enterprise Application Grid and Tangosol Coherence are examples of distributed memory caches. They are software-based caching solutions, as opposed to appliance-based caching solutions (we'll talk more about these later). It's easier to scale these types of solutions across the enterprise because you don't need to buy additional appliances. They can plug easily into distributed grid environments since they essentially provide a proxy (and a well-performing one at that!) to the underlying remote data. Most of them work within a fairly heterogeneous environment of application servers, databases and storage.

The downside to using this technology is that the cache needs to be managed and administered because it's a separate entity that is another component in the distributed computing environment.

Furthermore, a key consideration in using this technology is that the access interface should be essentially the same as it would be to the underlying data store. You want it to be transparent, whether an application or user is accessing the actual data or a cached version.

Data transformation

Data transformation may not seem like a core technology, but it is vitally important to providing true virtualization of underlying data sources.

Data transformation is the process of aligning and consolidating information from multiple systems and enriching the information to make it useful for new applications. For grids, it is especially important because of the potentially wide variety of data sources across a typical grid and because applications or users may request data to be provided in a wide variety of formats.

To truly virtualize information across a grid, it is likely that data transformation is going to have to occur. In fact, in typical extract-transformation-load (ETL) projects, data transformation consumes 45 percent of the project effort.

Think about that. It's easy to imagine, when you consider the effort any typical application spends manipulating and massaging data from one format to another. Products like WebSphere DataStage and DataStage Tx are designed to automate and manage the data-transformation effort. They are well worth the investment.

Approaches to consider

Now let's look at some approaches to consider, including:

  • Regional data centers.
  • Partitioning data.
  • Wide area file systems.

Regional data centers

One practical approach that helps mitigate access to distributed information is to co-locate data and computational at a physical location. Large users of grid computing, such as pharmaceutical and electronics companies, support multiple regional data centers or grids worldwide.

The grids at these data centers might have thousands of processors and typically a highly optimized storage area network infrastructure supporting fast access to the underlying physical data. Users then use a thin-client scheme to submit work and receive results.

In a sense, these companies are avoiding distributed data-access concerns entirely. Utilizing technologies such as caching and replication is less of a concern. The downside, however, is that they don't truly have a virtualized worldwide grid (in that users have to know where they are going for resources), but it is a very effective approach to minimizing data-access and transfer issues.

Partitioning data

Partitioning data according to some scheme or layout is an architectural approach to improving overall system throughput and performance that has been used for a long time and is applicable to distributed grids.

Usually, the partitioning layout creates mutually exclusive sets of data that certain resources in the grid infrastructure are dedicated to. Requests for particular information are then routed to the appropriate resources. This has the advantage that optimizations can take place between partitioned data and their associated resources resulting in significant data throughput improvements.

The downside of this approach is that it requires knowledge of the structure of data in order to effectively partition it. You pay the price of reduced flexibility with significant gains in data access and throughput.

Wide area file systems

Wide area file systems (WAFS) comprise another class of technologies that have recently seen increased momentum in the industry. Cisco Systems Inc., Brocade Communications Systems Inc., Tacit Networks Inc. and Riverbed Technology are companies with products in this space.

WAFS products are designed to optimize file access across a WAN from a data center to remote locations. Users install WAFS appliances at the central data center and at each of the remote locations. Thus, across the WAN, access to files in the data center is managed by WAFS services.

The way wide area file systems work is by cutting down on the nonessential traffic (the chattiness) between the data center and remote locations; they also employ intelligent caching. The downside of these appliances is that they are a hardware solution, so for each remote location, a new appliance is required. Also, they only work for files, but because a majority of data in a grid consists in a file format, this is somewhat less of a concern. Used appropriately, WAFS products can improve file data access in distributed grid environments.

Trends and directions

Let's take a look at some emerging trends that promise to address some of the data-access and performance challenges in grid environments.

Own your own network (fiber)

An emerging trend that might make sense for large, global users of grids across multiple locations is to look at purchasing dedicated network fiber, rather than contract with a service provider for networking services. With the bandwidth and capacity profile of optical fiber, this approach to addressing grid data challenges might seem akin to taking the sledgehammer out. But if you look at your network strategically and with the decreasing costs of optical fiber, it might make sense to own this important asset. Also, because companies are no longer reliant on network service providers, new network services can be provisioned in hours and days, not weeks and months.

For many users, the return on investment is worthwhile. “I think it's a strategy that companies need to look at,” said George Surdu, director of infrastructure at Ford Motor Co. “They need to work through the business case themselves. But I don't believe we are the first ones to think of doing this, and I'm sure we won't be the last.” The downside here comes with the management costs associated with operating your own network.

Data-aware scheduling

Workload schedulers, an important component of grids, traditionally have not taken into account data location when placing jobs for execution across distributed grids. Certainly executing work on resources that offer the best data-access and performance characteristics makes sense. So an emerging trend we are starting to see is schedulers taking into account the location of required data. This requires some way of associating required data (either input or output) and its location with a given executable or job.

Network-aware applications

A trend championed by Larry Smarr at UC San Diego is the concept of network-aware applications. “[The] network-aware application is a promising new concept [in] which applications are aware of network conditions and, thus, can adapt to the varying environment to achieve acceptable and predictable performance.”

As the characteristics of networks change in a distributed grid, placing network-aware intelligence in applications makes sense. The issue here is the programming model and the expectation that a broad set of application developers could easily integrate network intelligence in their applications.

In conclusion

In this article, we've presented a survey of technologies, approaches, and trends applicable to improving data-access performance and throughput in grid environments. It is essential to think how to apply these in the earliest stages of grid conception since data issues can quickly become a significant inhibitor to realizing success with grid computing. By doing so, data movement, access and throughput concerns can be mitigated.

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!

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion XL — were added to the benchmark suite as MLPerf continues 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 power it brings to artificial intelligence.  Nvidia's DGX Read more…

Call for Participation in Workshop on Potential NSF CISE Quantum Initiative

March 26, 2024

Editor’s Note: Next month there will be a workshop to discuss what a quantum initiative led by NSF’s Computer, Information Science and Engineering (CISE) directorate could entail. The details are posted below in a Ca Read more…

Waseda U. Researchers Reports New Quantum Algorithm for Speeding Optimization

March 25, 2024

Optimization problems cover a wide range of applications and are often cited as good candidates for quantum computing. However, the execution time for constrained combinatorial optimization applications on quantum device Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at the network layer threatens to make bigger and brawnier pro Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HBM3E memory as well as the the ability to train 1 trillion pa Read more…

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion 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…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HB Read more…

Nvidia Looks to Accelerate GenAI Adoption with NIM

March 19, 2024

Today at the GPU Technology Conference, Nvidia launched a new offering aimed at helping customers quickly deploy their generative AI applications in a secure, s Read more…

The Generative AI Future Is Now, Nvidia’s Huang Says

March 19, 2024

We are in the early days of a transformative shift in how business gets done thanks to the advent of generative AI, according to Nvidia CEO and cofounder Jensen 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…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. 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…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a 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…

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…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

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…

Shutterstock 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t 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…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

Intel Won’t Have a Xeon Max Chip with New Emerald Rapids CPU

December 14, 2023

As expected, Intel officially announced its 5th generation Xeon server chips codenamed Emerald Rapids at an event in New York City, where the focus was really o Read more…

IBM Quantum Summit: Two New QPUs, Upgraded Qiskit, 10-year Roadmap and More

December 4, 2023

IBM kicks off its annual Quantum Summit today and will announce a broad range of advances including its much-anticipated 1121-qubit Condor QPU, a smaller 133-qu Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire