TakTuk appear as a base tool for Grid5000/FutureGrid in the ANL MCS Technical Memo 326, "Supporting Experimental Computer Science".
TakTuk 3.7 series introduce major new features for deployment structuration and synchronization. The deployment groups enable the deployment of independent groups of machines with distinct deployment options and independent work stealing while still connecting all the machines to the same TakTuk network. Gateways support enable the use of machines as intermediate communication nodes that do not execute anything. Processes control is now more precise with the possibility of targeting taktuk commands to specific processes. Finally, introduced in 3.7.2, wait commands enable various flavors of synchronization between successive commands.
Starting with TakTuk 3.6.2, the new TakTuk::Pilot Perl module is included with the TakTuk distribution. This module ease the use of TakTuk from within a Perl script: it manages TakTuk execution, communication with the running TakTuk instance and I/O demultiplexing and decoding. This last feature is the most important as it allows applications to easily parse all the information outputed by TakTuk.
TakTuk has been used during the fifth GRIDS@work plugtest to deploy and execute a parallel application solving financial mathematics problem (assets pricing) and written using the KAAPI parallel programming environment. As in last year contest, application has been deployed on more than 3600 processor cores belonging to both Grid5000 (france) and Intrigger (japan). The time to complete the deployment of the application was only a few minutes (including the application initialization, TakTuk itself deployed processes in less than a minute). During this plugtest, TakTuk also served to KAAPI as a communication mean between Grid5000 and Intrigger.
Together with KAAPI, TakTuk earned the first prize from the plugtest jury as the fastest solution and the one that computed the largest number of solutions.
TakTuk has been used during the fourth GRIDS@work plugtest to deploy and execute a parallel application solving de Nqueens problem and written using the KAAPI parallel programming environment. The application has been deployed on more than 3600 processor cores. The time to complete the deployment of the application was only a few minutes (including the application initialization, TakTuk itself deployed processes in less than a minute).
Together with KAAPI, TakTuk earned the first prize from the plugtest jury as the fastest solution and the one that computed the largest number of solutions. All the details can be found on the Thierry Gautier's page
kanif, a TakTuk wrapper for cluster management and administration is out. This wrapper mimics syntax and behavior of well-known cluster tools C3, pdsh and dsh. It is designed to ease administrators tasks that deal with regular parallel machines.
TakTuk has been accepted as a part of the GNU/Linux Debian distribution (unstable branch, see this link). I sincerely acknowledge Lucas Nussbaum for his help regarding debian packaging, documentation proofreading and TakTuk testing.
A TakTuk user guide is now available for download in section "Documentation". The intend of this guide is to present main TakTuk characteristics and capabilities in a didactic way. It present in a progressive manner both performance and functional aspect of the tool. I hope it will be useful.
January, 26th 2007, TakTuk 3.0 is released. This stable release is the first official release of the new engine written in Perl. This new engine, compared to former ones written in C++, allows TakTuk to propagate its own code to remote hosts which eliminates the need to install TakTuk on all remote hosts. Overall the new engine is more customizable and now offer a basic communication layer.
TakTuk has been used during the third GRIDS@work Grid5000 plugtest to deploy and execute a parallel application solving de Nqueens problem and written using the KAAPI parallel programming environment. The application has been deployed on more than 1400 processors (over 700 dual processor machines) 12 successive times. The average time to complete the deployment of the application was less than 20s (estimated from the start time measured by the application itself).
Together with KAAPI, TakTuk earned the special prize from the plugtest jury as the fastest solution and the one that computed the largest number of solutions. All the details can be found on the plugtest results presentation and on Thierry Gautier's page