Research / Main applied results / Modern computer and information technologies in science and education
Modern computer and information technologies in science and education
The UARNet project, – Ukrainian Academic and Research Network, – was started in ICMP of NAS of Ukraine in 1992; and in February 1993 the first connection to the global Internet in the history of Ukraine through non-commutated communication lines was done as a part of that project. Later on, in 1995, a separate laboratory of information technologies and computer networks was created, which acts as a separate state enterprise "Scientific Telecommunication Center "Ukrainian Academic and Research Network" of ICMP of NAS of Ukraine starting from 1999.
In 2002 the first computer cluster in NAS of Ukraine was launched, it had 16 CPUs and computational performance of 12.5 Gflops. After being upgraded in 2004, 2006, and 2008 it is still one of the biggest in Ukraine. In December 2006 the ICMP cluster was built up with 17 new nodes based on dual-core 64-bit CPUs Xeon 5130 2 GHz and 4Gb memory. This had increased the tested performance up to 310 Gflops. In 2008 some of the dual-core CPUs were replaced with the quad-core CPUs, and the RAM was quadrupled. As the result, there had become totally 116 computational cores and 200Gb of RAM. All the nodes are connected with the Infiniband high-speed network giving the 700MB/s bandwidth in both directions at the same time. Thus, testing performance has risen to 640Gflops, and peak performance – to 928Gflops. Besides the computational power the computer cluster also needs the data storage system. The latter was implemented with a hard disk array linked with RAID-5 and using Lustre network file system, constituting a data storage server of 3TB volume.
The Rocks 4.2 Linux is used as the cluster operating system; and the SGE task control system is used to launch the calculations and control the computational power sharing. The Intel and Portland Group compiler packages are used on the ICMP cluster, the OpenMPI library is used to develop and run the parallel programs.
The cluster is connected to the power source via an uninterruptible power supply, constructed accordingly to computer system parameters. The system power analysis shows that it can run during 10-15 minutes after the external source is off, which gives enough time to gather the data from computational nodes and shut down the control node correctly. The cluster is in a room having all the needed ventilation systems and providing good conditions for computing.
The cluster setup, the installation and configuration of software and everyday maintenance is carried out by a group researchers from ICMP (T.Patsahan and O.Fareniuk). The cluster can be accessed directly from the workplaces in the institute or, – using secure connection, – from the Internet.
The high computational productivity of the cluster allows for using it to solve nationwide-level problems, which need the parallel software packages and large computer power in the first place. It is used today for the solution of complex interdisciplinary prediction problems, such as Chernobyl-posed problems. Now the ICMP cluster is used primarily for the simulation of physical and chemical processes in condensed matter by means of classical and ab initio molecular dynamics.
The cluster was connected to the Ukrainian GRID network in 2007. The constant modernization and development of effective simulation methods is directed towards the creation of a leading GRID node and refinement of Institute's position in the field.
ICMP focuses much of its efforts on having a highly qualified personnel to work with high-performance supercomputers, ranging from students to young scientists. As an example, the cluster is used to teach the basics of parallel programming and computer simulations to the students of local universities. To bring these ideas into being, the Institute has funded the creation of computer laboratory for students with direct access to the cluster, the appropriate study plans were developed, and students of Lviv universities and Institute researchers are taught the basics of parallel software packages (T.Bryk, T.Patsahan, I.Klevets).