Multicore NZ

January 23, 2012

Article: “The Memory Wall is ending multicore scaling”

Filed under: High Performance Computing, Integration and Services, Multicore — multicoreblog @ 8:58 am

From this article  at Electronic Design: “Multicore processors dominate today’s computing landscape. Multicore chips are found in platforms as diverse as Apple’s iPad and the Fujitsu K supercomputer. In 2005, as power consumption limited single-core CPU clock rates to about 3 GHz, Intel introduced the two-core Core 2 Duo. Since then, multicore CPUs and graphics processing units (GPUs) have dominated computer architectures. Integrating more cores per socket has become the way that processors can continue to exploit Moore’s law.”

“But a funny thing happened on the way to the multicore forum: processor utilization began to decrease. At first glance, Intel Sandy Bridge servers, with eight 3-GHz cores, and the Nvidia Fermi GPU, featuring 512 floating-point engines, seem to offer linearly improved multicore goodness.”

“But a worrying trend has emerged in supercomputing, which deploys thousands of multicore CPU and GPU sockets for big data applications, foreshadowing severe problems with multicore. As a percentage of peak mega-floating-point operations per second (Mflops), today’s supercomputers are less than 10% utilized. The reason is simple: input-output (I/O) has not kept pace with multicore millions of instructions per second (MIPS).”

Interesting.

 

 

September 1, 2011

Multicore World 2012 in Wellington, New Zealand

Filed under: Uncategorized — multicoreblog @ 9:34 pm

Open Parallel presents THE Multicore Conference.
Multicore World 2012 brings together Industry, Academia and Developers Communities to discuss the latest developments in Multicore Software and Hardware and its applications.
In Wellington, NZ. 27-28 March 2012
www.MulticoreWorld.com

June 15, 2011

Sun-Oracle SPARC five year plan

Filed under: Uncategorized — multicoreblog @ 9:41 am

From Oracle magazine (May 2011) we can read about how Oracle is looking at the good old SPARC architecture. The article presents a five year plan in these terms:

Five-Year Plan

When Oracle announced its SPARC roadmap at Oracle OpenWorld 2010, John Fowler, executive vice president of systems at Oracle, revealed a five-year trajectory for Oracle’s SPARC servers that included 4 times the number of cores between 2010 and 2015, 32 times the number of threads, 16 times the memory capacity, 40 times the number of transactions per minute, and 10 times the number of Java operations per second.

 

“Core to the server design element for Sun for many years has been SPARC, the first volume 64-bit processor. It’s about building mission-critical, high-performance systems for the enterprise,” said Fowler during his Oracle OpenWorld keynote. “We’re committing publicly . . . to at least double application performance every other year,” he added, noting that this commitment comes with binary investment protection for customers, who are often “picking . . . business applications and running them for many years, expecting performance improvements throughout.”

May 5, 2011

Parallel PHP: Open Parallel and Intel

Filed under: Multicore, Open Source, Parallel Programming — multicoreblog @ 9:42 pm

James Reinders, Chief Software Evangelist of Intel, posted this blog at Intel

http://software.intel.com/en-us/blogs/2011/05/04/parallel-php-hiphop-using-tbb-kiwi-style/

I’ve been chatting with a small group of dedicated fans of Intel Threading Building Blocks (TBB)  in New Zealand.  They’ve been looking at adding parallelism, using TBB, to WordPress, PHP, HipHop, Perl, and other open source projects.  They have published their code and some interesting results.  They have a web site http://openparallel.comexplaining some of their work.

The PHP (HipHip) project, using TBB, is hosted at https://github.com/openparallel/hiphop-php

Their PHP wrapper has been primarily implemented as an extension for the HipHop PHP compiler. This provides a thread-safe compiled implementation of PHP 5.2 with a fairly comprehensive set of libraries.

Their first version includes:

parallel_for / parallel_for_array - provides the TBB parallel_for functionality with PHP arrays for input and output.

concurrent_vector – this PHP extension class wraps the TBB concurrent_vector and provides thread-safe access to a vector type collection.

concurrent_hash_map – this PHP extension class wraps the TBB concurrent_hash_map and provides thread-safe access to a hash collection keyed on any PHP type.

concurrent_globals - this function provides thread-safe, read-only access to the PHP global variables. (These are normally thread-local in HipHop, where a thread is typically associated with a web request).
Some PHPDoc documentation has been produced for these functions and ore is being worked on. They are also looking at conventional PHP extensions to enable application developers to run an application that uses TBB extensions in the conventional PHP interpreter for development and testing purposes. In this case, all operations will execute on a single thread. This would support single source.

They presented much of their early work in January at a conference in Australia. I couldn’t swing visiting there myself, so I’ve resigned myself to occassional emails and phone calls to catch-up with them.

Here’s what they’ve done that I’ve spoken with them about:

They made slight modifications to WordPress, to use a TBB-enabled HipHop they created – and they had VERY impressive results. The bigger surprise was the huge drop in memory footprint. They have theories why, but this seems to have been the leading reason for the higher performance. Sometime algorithm changes improve performance for unexpected reasons!  You can read their two short white papers on what they did:

http://openparallel.wordpress.com/2010/11/01/tbb-in-wordpress-%E2%80%93-white-paper/

and

http://openparallel.wordpress.com/2010/11/01/wordpress-on-hiphop-white-paper/

I know they interested in hearing from developers of like mind (wanting to add parallelism to open source projects) – but they really enjoy talking with projects that want help improving performance. Given their results with PHP/HipHop so far, they would seem to be worth contacting for such work.

Parallelism is worth adding in many places.  It’s fun to see the results with PHP so far!

I know they are working on Perl too… I’ll catch up with them on that work and write a blog next week with what I find.

March 13, 2011

When will we see applications for multicore systems?

Filed under: High Performance Computing, Multicore, Parallel Programming — multicoreblog @ 10:23 pm

Keshav Pingali, a computer scientist at the University of Texas in Austin, is working with IBM under the auspices of Open Collaborative Research to develop the programming language that will give programmers the tools to write multicore-compatible code

Listen to Keshav’s podcast and read the transcript here

February 9, 2011

SuperComputer to be used for Research in Agriculture

The Centre for Development of Advanced Computing C-DAC with ten centres in major Indian cities,”is now assisting the Indian Council of Agricultural Research (ICAR) in establishing a national agricultural bioinformatics grid”

This initiative, the first of its kind in India, will help scientists enhance agricultural productivity and also address problems like food security. As part of the project, a three-day training-cum-workshop programme on ‘Parallel and High Performance Computing’ began on Monday 7.

The workshop will provide an insight into the different aspects of high performance computing (HPC) with the goal of capability building in solving complex problems in agriculture and biotechnology. Speaking to DNA, Goldi Misra, group coordinator and head, HPC Solutions Group, C-DAC, said the use of HPC would help scientists address the problem of food scarcity at the grass-root level. Full article.

January 31, 2011

“Reinventing Computing for Parallelism”

Filed under: Multicore, Parallel Programming — multicoreblog @ 11:06 pm

An article of general interest in PC World this week about “Radical Design for multicore” can be read here

The complete article “Using Simple Abstraction to Reinvent Computing for Parallelism” can be read here

“The ICE abstraction may take CS from serial (single-core) computing to effective parallel (many-core) computing.”

Uzi Vishkin – Communications of the ACM
Vol. 54 No. 1, Pages 75-85 (January 2011)

 

December 7, 2010

Schedule of the II Miniconference in Multicore and Parallel Computing, part of LCA2011

Filed under: Uncategorized — multicoreblog @ 12:31 am

With keynotes of Paul McKenney (Linux CTO of IBM) and Vint Cerf (Chief Internet Evangelist and VP of Google), the II Open Source Software, Multicore and Parallel Computing miniconference, will be in Brisbane, Australia on Tuesday 25 of January 2011

 

The schedule is available now.

Look forward to see you there!

Nicolás Erdödy

Miniconference Organiser – Multicore & Parallel Computing

LCA2011 – Brisbane, Australia

http://MulticoreLCA.wordpress.com/

 

October 20, 2010

Linux Scalability to Many Cores

Filed under: Multicore, Parallel Programming — multicoreblog @ 3:11 am

An Analysis of Linux Scalability to Many Cores is the title of a paper published by MIT. It discusses the scalability bottlenecks in a recent Linux kernel running in a 48 cores computer. A brief article about the paper is here

September 9, 2010

Google Code University

Filed under: Education and Training, Multicore, Parallel Programming — multicoreblog @ 3:39 am

With the headline

“One of the most important recent developments in computing is the growth in distributed and parallel applications.”,

Google presents an interesting list of very recent resources that is worth visiting

http://code.google.com/edu/parallel/index.html

Older Posts »

Theme: Shocking Blue Green. Blog at WordPress.com.

Follow

Get every new post delivered to your Inbox.