Thread locals

The whole database access code is surrounded by a try-finally block. There is no catch because the exceptions are rethrown by the code. However, the finally block is necessary because the ResultSet, Statement, and Connection objects must be closed properly whether an exception is thrown, or otherwise there will be database resource leaks. The database cleanup is implemented in a Util class. The initialization of the Connection object is purposely left out from the code. It can be either from a connection pool or from a Thread Local variable that is set earlier in the method call stack. Thread Local will be discussed further in Chapter 15. 

Production of carbon electrodes is a capital-intensive business. Two suppHers dominate the prebaked market. Carbon paste producers are more numerous and tend to serve local markets. There is no international standard for the threaded joints on carbon electrodes. Manufacturers of straight pin carbon electrodes have followed the physical specifications adopted for graphite electrodes (37). Unified standards do not exist for pinless joints resulting in limited interchangeability among brands. Electrode diameters are offered in both English and metric sizes with no restrictions on new or unique diameters. 

Figure 11.32 Severe localized metal loss on bolt threads and on nuts.

Hybrid OpenMP/MPI parallelization is used to compute Hv. The multithreaded OpenMP library is used for parallelization within each node, while MPI is used for communication between nodes. The subroutine fipsi that computes Hv contains two arrays of the same size the results of the portion of the calculation local to a node are stored in hps local, while the results of the portion of the calculation that requires communication are stored in hpsbujf. On each node, a dedicated (master) OpenMP thread performs all the communication, while other threads perform local parts of the work. In general, one uses as many threads as there are processors on a given node. The following scheme presents an overview of the execution of subroutine, hpsi ... 

Only the master thread performs communication, while all other threads perform the local part at the same time, leading to effective overlap of communication and computation. Once both local and communication parts are finished, array hpsbujf is added to the array hps local. The barrier above denotes the place where all the threads are synchronized. It is necessary to ensure that addition of hpsbujf io hps local will not start before both local and communication parts are finished. 

The master thread performs all the intemode communication using MPI. Only the action of the kinetic energy operators, Tr and Tr implemented with the DFFD approach, require such communication. While the master thread is executing the communication part and storing its results in hpsbuff, other threads perform local work storing results in the main array hps local. The use of the two separate arrays is needed to avoid having to synchronize the threads. Moreover, if the communication part finishes before the local part, the master thread joins the other threads in the computation of the local part. 

The shared memory OpenMP library is used for parallelization within each node. The evaluation of the action of potential energy, rotational kinetic energy, and T2 kinetic energy are local to each node. These local calculations are performed with the help of a task farm. Each thread dynamically obtains a triple (/2, il, ir) of radial indices and performs evaluation of first the kinetic energy and then the potential energy contribution to hps local( , i2, il, ir) for all rotational indices. 

Once both local and communication parts of the Hamiltonian evaluation are finished, we add the buffer hpsbujf contmmng results of the communication part to the main array hps local. However, it is necessary to ensure that this addition will not start before both local and communication parts are finished. A barrier directive forces all the threads to synchronize. A thread that has reached the barrier will not resume execution until all other threads have reached it too. 

A random coil is clearly a three-dimensional object when looked at from long distance. Locally, however, it resembles more a one-dimensional thread. Therefore it is sensible to describe the coil by a fractal dimension that lies closer to 1 (for other architectures somewhere between 1 and 3). Such disordered objects are called fractals [101,102]. 

---