This particular application is using one thread per user, so theoretically, if one has an 8-core machine then 8 concurrent users should fully utilize the machine giving 8x speedup compared to a sequential run. Recently I have got a workload that could not scale beyond a few cores.

See Page 11 of this document for details. Locks in the Linux page fault handlers have been discussed here. Then I used my best friend VTune™ to see what are the top hot functions in the code. Virtual to physical address mapping entries are created later upon a page fault in a lazy manner, when the virtual memory pages are touched by the process (i. I have located the piece of code that called that memset function. This was a real surprise for me, because it was known that the application did not use memory mapped files. The Intel Thread Profiler had no chance to show the possible locks in page fault handlers because these handlers are executed in Linux kernel (Intel Thread Profil er operates in user space). Then I had understood where these mmaps/munmaps are coming from. with memset) for the first time. These were mmap, munmap and memset. It was an initialization of recently allocated memory blocks.

e. The Lin ux mmap mechanism allocates new memory by assigning page file space to virtual address space of the app. It tu rned out that glibc memory alloc function (malloc) is using memory mapping mechanisms of Linux to allocate bigger chunks of memory (at least for my Linux distribution with the stock kernel). This is why there were a huge number of context switches (shown by the vmstat utility); page faults cause switches from user space to kernel space.

I have seen concurrent workloads that were speedup by a few percent only and I find this still encouraging to give Intel TBB allocator a try.

This is of course no guarantee that any arbitrary multithreaded application will profit from Intel TBB allocator at this scale.

so. 2:$LD_PRELOAD
run_your_app

You can verify if the allocator library is loaded using “pmap app_process_id | grep tbb”
2 that also supports large object allocation efficiently. One should just define some environment variables to use it for a legacy application:

export LD_LIBRARY_PATH=absolute_path_to_tbb_libs:$LD_LIBRARY_PATH

export LD_PRELOAD=libtbbmalloc_proxy. so. I knew that Intel Threading Building Blocks (Intel TBB) has a scalable memory allocator, and recently a drop-in version has become available in version 2. 2:libtbbmalloc.

It was the memory management that did not scale. It looked like the problem had been reproduced.

Now come the results with Intel TBB malloc lib:
1 thread: 6 seconds runtime, at most 1 core is busy
2 threads: 6 seconds runtime, at most 2 cores are busy
4 threads: 10 seconds runtime, at mos t 4 cores are busy
8 threads: 14 seconds runtime, at most 8 cores are busy

feedback and questions are very welcome.

To verify my hunch that the memory management via standard malloc is not scaling as one could hope for, I wrote a simple program that did just that in every thread:

void * worker(void *) {
int iterations = 300000;
while(iterations– ) {
void * p = malloc(size);
memset(p,0,size);
free(p);
}
return NULL;
}

My test spawned x worker threads and then waited until they are all finished.

Roman

or exclusive locks? I have used Int el Thread Profiler which has hel ped me a lot previously. It means it should scale. It has shown a perfect picture: with 8 users there were 8 threads busy, halt ing very rarely on a few locks. It was known to me that the relevant portion of the code could be locked with a read-write lock. I talked to the developers of this application. The queries were read-only. So, it should scale!

Too much sequential code in the processing? Then spent some time looking at the source code and guessing what the reason could be.

Looking at the vmstat output one could observe that the number of context switches was back to normal values. For this small test routine the Intel TBB malloc speedup was amazing 23x My original workload – ran about 7x faster utilizing up to 8 cores with this new allocator.

Here are results: for 200KB blocks (size=200*1024):
1 threa d: 21 seconds runtime, at m ost 1 core is busy (via top utility)
2 threads 53 seconds runtime, at most 1. 4 cores are busy
4 threads: 105 seconds runtime, at most 1. 8 cores are busy
8 threads: 325 seconds runtime, at most 2 cores are busy

It did not happen. At most two cores have been utilized, the query throughput speedup was even smaller.

VTune is a trademark of Intel Corporation in the U. and other countries

S.

software.intel.com

During day three of Game Developers Conference 2009 we had another wow! moment which had to do with the brand-new development tool based on CryEngine 3 being presented at the Crytek booth.

In addition to the real-time changes single updates on a dedicated game level will be automatically executed even if one of the platforms crashes during the design work- one more huge advantage for game developers.

With CryEngine 3 you can develop new games for all three platforms – PC, PlayStation and XBox – at the same time. This means that all changes within the same game level can be tested at real-time for all platforms. And this will dramatically reduce development time – and cos t.

But also the integr ation of external tools like Photoshop into Crytek’s developing environment is an amazing feature: You can manipulate existing textures with your picture editing tool and change the texture in the game level easily. It was probably never easier to be a game designer. Even the design of new objects like islands, rocks and so forth works fast and efficiently.

So let’s look forward to CryEngine 3 based games, although the differences won’t be so dramatically compared to CryEngine 2.

And as we’ve heard by Crytek’s CEO Cevat Yetil we still have to be a little more patient as the next generation of CryEngine will only show up in 2012.

software.intel.com

Together, they represent a good chunk of the industry and their sheer size and reach should establish WiGig as a new interoperable standard.

WiGig is a device to device (p2p) network and does not require a central HUB or router that could easily turn into a congestion point. Because it is different, its goal is not to replace WiFi or Bluetooth, although you can imagine some overlapping there. WiGig’s weak point is that its signal can be easily stopped or strongly weakened by obstacles like walls. WiGig has been designed to provide ultra-fast data transfers (up to 6GB/sec, theoretically) over relatively short distances (about 20 yards, without walls).

At this point, the specifications will be done by December 2009 but you can bet that teams are ready at work on the hardware implementations. This is critical is the success of WiFi is built on a worldwide interoperability. WiGig is based on the 60Ghz frequency, which is available worldwide. Any company can join the alliance, which has learned greatly from the success of WiFi.

” Whatever the outcome is, we look forward to this revolutionary way Apple has updated their iPod nano, wondering how many more improvements can be made that fall under the quantum leap category in the near future. Recent reports have pointed toward the next generation iPod nano (which will be its 5th iteration) that will feature an integrated digital camera as well as a more compact, circular scroll wheel that has a way to detect “input gestures that traverse the center of the scroll wheel and to detect multi-touch input. Apple’s famous Click Wheels on their iPod range could soon get a similar multi-touch treatment that is found on their iPhone, iPod touch and Macbook lines.

With the proliferation of integrated webcam, there’s no doubt that video calls have been on the rise. While that setup is good enough for most people, the Dualpix HD720 offers the option of putting the webcam on the side, and orient it in more directions (front + left/right). $59 in many online retailers. That said, the image quality of integrated cameras is quite low and there’s an assumption that you will stand right in front of the display. Secondly, the additional room allows for the use of a better lens and image processor (in theory at least, we have not tried it for ourselves). Product page

And ‘your first time here?

The application implements the main mixing console, while being loaded as a plug separate the sequencer, the form of audio input and other features.
For this reason, Kristal Audio Engine is a good choice for those who again wish the first steps into the world of home recording without turning to a commercial package. Kristal Audio Engine multi-track recorder is a good modular architecture.

This freeware program can handle 16 audio tracks, with sample rate from 44.1 to 192 kHz.

Despite not being updated since 2004, this