Showing posts with label High Performance Computing. Show all posts
Showing posts with label High Performance Computing. Show all posts
Monday, January 20, 2025
Tuesday, April 28, 2020
Monday, March 23, 2020
Sunday, March 28, 2010
Concurrency in Java 5
J2SE 5 very much extends and expands the concurrency model of Java. The very useful concurrency communication classes such as (bounded and unbounded) queues are introduced.
One of the best parts of new improved Java concurrency model is the new java.util.concurrent.BlockingQueue interface.
LinkedBlockingQueues and ArrayBlockingQueues do away with wait(), notify() and notifyAll() style of coding. Add to that Synchronized queues, which are simply amazing, as well as DelayQueue classes and PriorityBlockingQueues and you have yourself a whole lot of threading fun!
Other useful classes are CountDownLatch, CyclicBarrier, Executor,
CachedThreadPools, FixedThreadPools, SingleThreadExecutors, ExecutorServices, Futures, and Callable.
I think it is a good idea to look into learning synchronous, asynchronous queues, and cached threading pools, and then delve into performance considerations for atomic, lock, and concurrent methods of class, method, and attribute concurrency.
Note that there are command line switches important to concurrent programming (-X*) and also that there Runtime classes designed to help avoid OutOfMemory heap problems.
Some of these exciting stuff is covered in the "Thinking in Java 4" by Bruce Eckel but that book does not concentrate on concurrent programming.
The book to read is:
Java Concurrency in Practice
by Brian Goetz, Tim Peierls, Joshua Bloch, Joseph Bowbeer, David Holmes, Doug Lea
One of the best parts of new improved Java concurrency model is the new java.util.concurrent.BlockingQueue interface.
LinkedBlockingQueues and ArrayBlockingQueues do away with wait(), notify() and notifyAll() style of coding. Add to that Synchronized queues, which are simply amazing, as well as DelayQueue classes and PriorityBlockingQueues and you have yourself a whole lot of threading fun!
Other useful classes are CountDownLatch, CyclicBarrier, Executor,
CachedThreadPools, FixedThreadPools, SingleThreadExecutors, ExecutorServices, Futures, and Callable.
I think it is a good idea to look into learning synchronous, asynchronous queues, and cached threading pools, and then delve into performance considerations for atomic, lock, and concurrent methods of class, method, and attribute concurrency.
Note that there are command line switches important to concurrent programming (-X*) and also that there Runtime classes designed to help avoid OutOfMemory heap problems.
Some of these exciting stuff is covered in the "Thinking in Java 4" by Bruce Eckel but that book does not concentrate on concurrent programming.
The book to read is:
Java Concurrency in Practice
by Brian Goetz, Tim Peierls, Joshua Bloch, Joseph Bowbeer, David Holmes, Doug Lea
Wednesday, January 20, 2010
Mitrion-C
Mitrionics AB (http://www.mitrionics.com/ ) has released Mitrion-C for writing parallel code. This is a new language that uses "C" syntax and is optimized for parallel compilation on FPGA as well as multi-core processors. See the Company announcement @ http://www.mitrionics.com/?page=view_release&itemid=9947&back=mitrionics-sc09-press-kit
The language details may be found @ http://www.science.uwaterloo.ca/~hmerz/FPGA/The%20Mitrion-C%20Programming%20Language.pdf
The language details may be found @ http://www.science.uwaterloo.ca/~hmerz/FPGA/The%20Mitrion-C%20Programming%20Language.pdf
Saturday, October 3, 2009
Threading References
From the article “Three Sides of Threading” by Claire Cates in the October 2009 issue of the Software Test & Performance Magazine:
Developing Multithreaded Applications: A Platform Consistent Approach @
http://cache-www.intel.com/cd/00/00/05/15/51534_developing_multithreaded_applications.pdf
Multithreading Programming Guide @ http://docs.sun.com/app/docs/doc/816-5137
Dr. Dobbs: Performance Analysis and Multicore Processors : March 30, 2006 @ http://www.ddj.com/dept/64bit/184417069
Dr Dobbs: Multitasking Alternatives and the Perils of Preemption: Sept 14, 2006 @ http://www.ddj.com/dept/embedded/193000965
Data Placement in Threaded Programs: from the Intel Software Network Threading Methodology: Principles and Practices @
http://cache-www.intel.com/cd/00/00/21/93/219349_threadingmethodology.pdf
Intel Software Developer Webinar Series, Fall 2007
W. Brown, R. Malveau, H. McCormick III, T. Mowbray, “AntiPatterns: Refactoring Software, Architectures, and Projects in Crisis,” John Wiley, New York (1998)
C. Smith, L.Williams, “More New Software AntiPatterns,” Proceedings of 2003 Computer Measurement Group
Developing Multithreaded Applications: A Platform Consistent Approach @
http://cache-www.intel.com/cd/00/00/05/15/51534_developing_multithreaded_applications.pdf
Multithreading Programming Guide @ http://docs.sun.com/app/docs/doc/816-5137
Dr. Dobbs: Performance Analysis and Multicore Processors : March 30, 2006 @ http://www.ddj.com/dept/64bit/184417069
Dr Dobbs: Multitasking Alternatives and the Perils of Preemption: Sept 14, 2006 @ http://www.ddj.com/dept/embedded/193000965
Data Placement in Threaded Programs: from the Intel Software Network Threading Methodology: Principles and Practices @
http://cache-www.intel.com/cd/00/00/21/93/219349_threadingmethodology.pdf
Intel Software Developer Webinar Series, Fall 2007
W. Brown, R. Malveau, H. McCormick III, T. Mowbray, “AntiPatterns: Refactoring Software, Architectures, and Projects in Crisis,” John Wiley, New York (1998)
C. Smith, L.Williams, “More New Software AntiPatterns,” Proceedings of 2003 Computer Measurement Group
Friday, September 18, 2009
RapidMind
RapidMind is a multi-core development platform that supports Streaming SIMD Extensions. It was recently bought by Intel. Learn more about it @ http://www.rapidmind.com/.
Cilk++
Cilk++ is three keywords and a runtime system that extend C++ to the realm of multicore (or parallel) programming. Cilk++ is platform-independent. It was recently bought by Intel. Learn more about it @ http://www.cilk.com/multicore-products/cilk-solution-overview/
Tuesday, July 7, 2009
Axum
New toys from Bill's boys er Steve's boys @ http://msdn.microsoft.com/en-us/devlabs/dd795202.aspx
Thursday, June 25, 2009
WhisperStation-PSC
WhisperStation-PSC is a personal supercomputer featuring Tesla GPUs and the CUDA SDK. It contains up to four Tesla C1060 GPUs, two AMD or Intel processors, 64 GB memory and optional RAID storage.
The supercomputer is designed for computationally-intensive OpenMP applications that employ hundreds of parallel threads to speed up floating point operations, such as industrial design, financial modeling, space exploration, CFD simulation and medical imaging.
Learn more @ www.microway.com
The supercomputer is designed for computationally-intensive OpenMP applications that employ hundreds of parallel threads to speed up floating point operations, such as industrial design, financial modeling, space exploration, CFD simulation and medical imaging.
Learn more @ www.microway.com
Saturday, June 13, 2009
HIPerWall
HIPerWall is a wall built of numerous high-definition monitors, each with its own imbedded computer for displaying standard and large (up to one gigabyte or larger) graphic images, high-definition (HD) digital movies, and streaming content from video cameras and other live feeds.
Built at Calit2 (California Institute for Telecommunications and Information Technology) at UC Irvine, the Highly Interactive Parallelized Display Wall (hence HIPerWall) has the ability to display 200 Megapixel images.
It is designed to visualize enormous data sets and allows viewers to see detail, with 100 dots per inch on the screens, while retaining the context of an overview by seeing surrounding data (also in high detail). This allows a group of scientists to collaborate, share detailed information, while still keeping the big picture.
Learn more @ http://hiperwall.calit2.uci.edu/
Built at Calit2 (California Institute for Telecommunications and Information Technology) at UC Irvine, the Highly Interactive Parallelized Display Wall (hence HIPerWall) has the ability to display 200 Megapixel images.
It is designed to visualize enormous data sets and allows viewers to see detail, with 100 dots per inch on the screens, while retaining the context of an overview by seeing surrounding data (also in high detail). This allows a group of scientists to collaborate, share detailed information, while still keeping the big picture.
Learn more @ http://hiperwall.calit2.uci.edu/
Thursday, June 11, 2009
More on Parallel Processing
I found the various comments posted at the GCN Government Computer News blog (http://gcn.com/Blogs/Tech-Blog/2009/06/New-parallel-processing-languages.aspx ) quite insightful.
Take a look!
Take a look!
Thursday, April 16, 2009
Tuesday, February 3, 2009
Approaches to Parallelism
A new model for writing concurrent applications is based on a structure called an actor. An actor is a computation entity whose primary actions are performing operations that are passed to it, passing data to other actors, and creating new actors.
These operations are entirely local; they have no effects on other actors. To affect other actors, an actor must pass a message to them, including the data they need or the new instructions.
This parallel-programming technique does not use variable but constants – like Java strings - as data cannot be unexpectedly changed by another thread.
Data changes among actors are done by sending messages to one another. Because of this built-in mutual exclusion, actors are a good match for parallelism
Erlang is the language with the greatest commercial acceptance that uses actor-like constructs. For Java, there is an actor framework called ActorFoundry. But, on the JVM, the best choice is the emerging language Scala, which provides support for traditional object-oriented-style programming as well.
This approach is similar to dataflow, a design that was first expounded in the 1960s. It too uses message passing and adds a built-in capability to monitor the relationship between two data items, such that if one changes, the other is automatically updated.
Pervasive Software is releasing DataRush Java library which has handled massive amounts of data in tests with only modest hardware platforms. The product has the ability to leverage data-flow across all the processor cores.
It seems that message-passing parallelism (think OpenMP) is likely to become more prominent during the next few years as a way to leverage the many cores in today’s PCs and servers.
These operations are entirely local; they have no effects on other actors. To affect other actors, an actor must pass a message to them, including the data they need or the new instructions.
This parallel-programming technique does not use variable but constants – like Java strings - as data cannot be unexpectedly changed by another thread.
Data changes among actors are done by sending messages to one another. Because of this built-in mutual exclusion, actors are a good match for parallelism
Erlang is the language with the greatest commercial acceptance that uses actor-like constructs. For Java, there is an actor framework called ActorFoundry. But, on the JVM, the best choice is the emerging language Scala, which provides support for traditional object-oriented-style programming as well.
This approach is similar to dataflow, a design that was first expounded in the 1960s. It too uses message passing and adds a built-in capability to monitor the relationship between two data items, such that if one changes, the other is automatically updated.
Pervasive Software is releasing DataRush Java library which has handled massive amounts of data in tests with only modest hardware platforms. The product has the ability to leverage data-flow across all the processor cores.
It seems that message-passing parallelism (think OpenMP) is likely to become more prominent during the next few years as a way to leverage the many cores in today’s PCs and servers.
Saturday, October 11, 2008
Amazon EC2 API
The Amazon Elastic Compute Cloud (Amazon EC2) web service API gives on the ability to execute arbitrary applications in the Amazon computing environment.
EC2 API has the momentum to become the x86 "chipset" of Cloud Computing.
Learn more about it @ http://docs.amazonwebservices.com/AmazonEC2/gsg/2006-06-26/
EC2 API has the momentum to become the x86 "chipset" of Cloud Computing.
Learn more about it @ http://docs.amazonwebservices.com/AmazonEC2/gsg/2006-06-26/
Tuesday, August 19, 2008
Free .Net Deadlock Debugging Tool
Corneliu, creator of the GUI debugging tool HawkEye, has a debugging tool that works against most .NET programs, without recompiling them. The follwoing are its features:
1. The tool does not require to have the code re-compiled in any way or form, with any external dependencies, nor reference any external library or have you modify your code to use any special type of locks inside your code.
2. It works on release builds with no PDB files.
3. It works on running processes or previously captured memory dumps.
4. It detects deadlocks across multiple threads and returns detailed call-stack and lock usage information.
5. It only detect deadlocks in which threads are actively waiting for locks acquired by other threads.
6. It does not detect the dining philosophers problem or deadlocks created in combination of time waits + wake/check + lock.
7. It has an external dependency on the cdb.exe (part of the the free Debugging Tools for Windows package from Microsoft).
8. It requires absolutely no installation. It an xcopy deployment.
9. And best of all it’s free (source code to be published soon)
The tool is discussed @ http://www.acorns.com.au/blog/?p=129 and may be obtained @ http://www.acorns.com.au/files/ACorns.Debugging.DeadlockTests.1.0.1.zip
1. The tool does not require to have the code re-compiled in any way or form, with any external dependencies, nor reference any external library or have you modify your code to use any special type of locks inside your code.
2. It works on release builds with no PDB files.
3. It works on running processes or previously captured memory dumps.
4. It detects deadlocks across multiple threads and returns detailed call-stack and lock usage information.
5. It only detect deadlocks in which threads are actively waiting for locks acquired by other threads.
6. It does not detect the dining philosophers problem or deadlocks created in combination of time waits + wake/check + lock.
7. It has an external dependency on the cdb.exe (part of the the free Debugging Tools for Windows package from Microsoft).
8. It requires absolutely no installation. It an xcopy deployment.
9. And best of all it’s free (source code to be published soon)
The tool is discussed @ http://www.acorns.com.au/blog/?p=129 and may be obtained @ http://www.acorns.com.au/files/ACorns.Debugging.DeadlockTests.1.0.1.zip
Saturday, July 26, 2008
Lilly Science Grid
Eli Lilly has made its Discovery IT platform, known internally as the Lilly Science Grid (LSG), open source.
LSG is a plug-in hosting and deployment framework that sits on top of Microsoft’s Composite Application Block. It is a rich client that requires .NET 2.0 or higher.
The framework simplifies the task of creating new plug-ins by providing a Visual Studio template from which developers can quickly learn and expand. Users can choose which applications and plug-ins to use within an integrated environment.
Find it @ http://sourceforge.net/projects/lsg
LSG is a plug-in hosting and deployment framework that sits on top of Microsoft’s Composite Application Block. It is a rich client that requires .NET 2.0 or higher.
The framework simplifies the task of creating new plug-ins by providing a Visual Studio template from which developers can quickly learn and expand. Users can choose which applications and plug-ins to use within an integrated environment.
Find it @ http://sourceforge.net/projects/lsg
Thursday, April 17, 2008
Actor Model of Concurrency
Erlang implements the Actor Model of concurrency. Learn more @ http://en.wikipedia.org/wiki/Actor_model and @ http://www.ccs.neu.edu/home/tov/erlang-talk/
AMD CTM
This is a free interface for interacting with AMD's GPU products (similar to NVDIA's CUDA). It is available @ http://ati.amd.com/companyinfo/researcher/resources.html
Thursday, April 10, 2008
NVIDIA Tesla C870
NVIDIA Tesla C870 is a Graphics Processor Unit (GPU) designed for high performance computing (HPC) solutions. It brings supercomputing power to any workstation or server and to standard, CPU-based server clusters.
Key elements include:
massively multi-threaded architecture with a 128-processor computing core
A C-language development environment for the GPU (called CUDA)
A suite of developer tools (C-compiler, debugger, performance profiler, optimized libraries)
Seamlessly fits into existing HPC environments
The CUDA programming guide is a good place to start. You can get it @ http://developer.nvidia.com/object/cuda.html
Then download the SDK. The sample code are a great learning resource.Also go through the UIUC course on parallel programming with special focus on CUDA @ http://courses.ece.uiuc.edu/ece498/al1/Syllabus.html
The SDK indicates simple ways to integrate C++ and CUDA
Key elements include:
massively multi-threaded architecture with a 128-processor computing core
A C-language development environment for the GPU (called CUDA)
A suite of developer tools (C-compiler, debugger, performance profiler, optimized libraries)
Seamlessly fits into existing HPC environments
The CUDA programming guide is a good place to start. You can get it @ http://developer.nvidia.com/object/cuda.html
Then download the SDK. The sample code are a great learning resource.Also go through the UIUC course on parallel programming with special focus on CUDA @ http://courses.ece.uiuc.edu/ece498/al1/Syllabus.html
The SDK indicates simple ways to integrate C++ and CUDA
Subscribe to:
Posts (Atom)
About Me
- Babak Makkinejad
- I had been a senior software developer working for HP and GM. I am interested in intelligent and scientific computing. I am passionate about computers as enablers for human imagination. The contents of this site are not in any way, shape, or form endorsed, approved, or otherwise authorized by HP, its subsidiaries, or its officers and shareholders.