Getting Serious About Transactional Memory

By Michael Feldman

January 12, 2007

The parallelization of computing, via multi-threading cores, multi-core processors and multi-processor systems is encouraging ever greater levels of application concurrency to take advantage of the proliferating CPUs. Multi-core processors, in particular, are fueling this phenomenon.

Beyond the dual- or quad-core domain, manufacturers are starting to build many-core chips. Examples include the Niagara T1 UltraSPARC from Sun Microsystems, which has 8 cores and can support 32 threads (the next-generation Rock processor will double this to 16 cores and 64 threads); Cavium Networks 16-core OCTEON MIPS64 processor for embedded applications; and Intel's Polaris prototype processor, which sports 80 cores and boasts a peak teraflop.

The Polaris prototype, which is part of Intel's terascale computing initiative, is motivating researchers there to take a hard look at a relatively new technology — transactional memory or TM, for short. Within the next ten years, the prospect of multi-core teraflop processors like Polaris and new application domains to use those processors will require vastly more parallel processing than ever before. Even incorporating relatively low levels of concurrency in today's applications is already challenging some of our best developers.

One of the nastiest concurrency problems has to do with keeping data thread-safe, that is maintaining global data integrity in the presence of parallel executing threads. Failure to keep data thread-safe leads to deadlocks, race conditions (data corruption) and priority inversion. Worse, because these types of problems are time-sensitive, they are often very hard to find during normal testing and in some cases go undetected until after the application is deployed.

The typical way to keep data thread-safe is to use global locks around objects that are being accessed by more than one thread. Locks provide a synchronization mechanism that blocks concurrent access of an object, preventing the data race condition. Seems simple enough. But there are a number of problems with this approach. Sometimes locks become dependent on each other, such that each thread is holding a lock the other thread needs. Or if a thread dies holding a lock it can block other dependent threads.

Even for correctly implemented locks, there's the issue of granularity. Coarse granularity protects larger data objects and uses fewer locks. But as the number of threads scales up, performance suffers. Finer granularity allows the programmer to protect smaller data items and gives better performance as long (as lock overhead is not overwhelming). It makes it possible, for example, to lock individual record components rather than the entire record structure. But finer granularity requires more complex algorithms and more locks, so it is often much more difficult to implement correctly.

Transactional memory to the rescue

Terascale computing, which relies on many-core parallelism, will be very difficult to develop. The current languages only provide low-level concurrency features. For Intel, terascale computing has become the prime motivator to improve software concurrency technology. The company's 80-core Polaris prototype will require much greater levels of application concurrency than today. And the scaling up of multi-core processors across time and product families will necessitate a solution that doesn't require reprogramming based on core count.

“How can the programmer write parallel code more effectively, that is, write robust code that doesn't have bugs, but still scales and benefits from the additional cores that each successive generation provides,” asks Ali-Reza Adl-Tabatabai, Intel Principal Engineer? “That's the big challenge that we're going after.”

To that end, Intel researchers are looking to transactional memory as one of the key technologies that will enable developers to write the terascale killer apps of the next decade. The attraction of TM is that is appears to solve the most annoying problems of global locks: application robustness and scalability. These attributes are especially important for the type of large-scale concurrency required by terascale applications.

Like locks, transactional memory is a construct for concurrency control that enables access to data shared by multiple threads. But unlike locks it is an optimistic model. It assumes that in most cases only a single thread will be contending for a given data item. A transaction is a high-level construct that executes reads and writes to data as an indivisible operation. From the application's point of view intermediate states are not visible to other successful transactions. So when a logical transaction is complete, the system verifies that other logical transactions haven't made changes to the same memory that would conflict with the first transaction. If they have, then the transaction is re-executed until it succeeds.

Intel's view is that TM should be encapsulated in language construct, and initially implemented in software. At some point, it may be useful to provide a hardware assist to provide better performance or functionality. But this is not necessarily the case.

Adl-Tabatabai says that dynamic memory garbage collection, a technology introduced about 50 years ago, may be a good analogy of how transactional memory will evolve. Initially, there were a number of techniques for implementing garbage collection in software. People thought that they really would like hardware support for this. But the software algorithms evolved and eventually made it into mainstream languages like Java. It turned out that hardware assistance wasn't really needed. In any case, once the language semantics of TM are defined, the software/hardware implementation should be transparent to the application developer.

Intel researchers have prototyped extensions to Java and C that incorporate transactional memory constructs. They've also investigated using various compiler optimizations for transactional memory implementations. Below are two trivial code examples that make data object 'x' thread-safe. The first uses explicit locking, the second uses the atomic block construct for transactional memory.

   lock(L); x++; unlock(L);

   atomic {x++;}

The atomic construct guarantees the enclosed operations will be safe from thread concurrency. The data transactions within the construct will either execute completely or have no effect until it is safe to do so. When the atomic block executes this happens as if in a single step in relation to the other threads. In other words, from the programmer's point of view, the transactions run in isolation.

The DARPA HPCS research languages (Fortress, X10, Chapel) for high productivity computing all provide atomic block construct in lieu of explicit lock synchronization. A few other investigators have incorporated TM into other research languages, but the final language to emerge from the DARPA HPCS program may be the first one to formally introduce it as a standard feature.

“The fact that the HPCS languages decided to provide atomic constructs rather than locking constructs shows that there's general consensus in the language design community that this is the way to go for concurrency control in future languages,” notes Adl-Tabatabai.

According to Adl-Tabatabai, The HPCS language effort represents a real step forward, since it will incorporate TM from the beginning. Retrofitting transactions into older languages that previously used locks can be problematic, since mixing explicit locking and implicit locking via a low-level implementation of TM may introduce conflicts when dealing with legacy code.

Scaling to advantage

On many applications, coarse-grained locking — putting locks around whole data structures — doesn't scale well. Threads that are concurrently accessing the same data structure must wait unnecessarily when they are reading or writing disjoint data within the structure. To increase performance, the developer must re-program the algorithm using fine-grained locking — putting global locks around individual data items in the data structures. Fine-grained locking allows different threads to concurrently access disjoint data in the same data structures. Transactional memory implicitly provides fine-grained locking, automatically providing the associated performance benefits.

Adl-Tabatabai describes an example using traditional fine-grained locking for a hash table. Using the Java 5 class libraries, a professor of computer science (who wrote a book on Java concurrency) was able to make the appropriate modifications to the hash table algorithms to incorporate fine-grained locking. The changes were subtle, yet quite complex. After two years of reviews by the Java standards committees, it was approved for inclusion into the Java class libraries.

“In general, doing this kind of coding is very difficult,” says Adl-Tabatabai. “You're not going to be able to get your average Joe Programmer to write code like this, and write it in a way that's error-free and doesn't introduce data races or deadlocks.”

Besides fine-grained locking, TM has the additional advantage of allowing for concurrent reads of the same data, which traditional locks cannot do. A special type of lock, called a reader-writer lock can be used to overcome this deficiency but this requires application code modifications. Transactional memory, on the other hand provides a systematic way for programmers to take advantage of these features.

With no existing base of software, how will TM get mainstreamed into applications? Jerry Bautista, Director, Microprocessor Technology Management at Intel says the evolution to multi-core architectures will create a need in the marketplace for technologies like transactional memory as the complexity of programming creates a new set of challenges.

“As we introduce more high performance hardware, there will be a demand generated in the programming community for ways to get around these common problems,” says Bautista. “People are already noting these challenges today. We're just trying to run ahead here.”

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!

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…

Nvidia Appoints Andy Grant as EMEA Director of Supercomputing, Higher Education, and AI

March 22, 2024

Nvidia recently appointed Andy Grant as Director, Supercomputing, Higher Education, and AI for Europe, the Middle East, and Africa (EMEA). With over 25 years of high-performance computing (HPC) experience, Grant brings a 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…

Houston We Have a Solution: Addressing the HPC and Tech Talent Gap

March 15, 2024

Generations of Houstonian teachers, counselors, and parents have either worked in the aerospace industry or know people who do - the prospect of entering the fi 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