We implement a high-order finite-element application, which performs the numerical simulation of seismic wave propagation resulting for instance from earthquakes at the scale of a continent or from active seismic acquisition experiments in the oil industry, on a large cluster of NVIDIA Tesla graphics cards using the CUDA programming environment and non-blocking message passing based on MPI.
We use mesh coloring to efficiently handle summation operations over degrees of freedom on an unstructured mesh, and non-blocking MPI messages in order to overlap the communications across the network and the data transfer to and from the device via PCIe with calculations on the GPU.
With this communication scheme based on non-blocking and persistent messages, a higher speedup than the one shown in Figure 10 could be expected.
Therefore, we propose a scheduling algorithm that may be performed before the execution of an application in order to compute a non-blocking schedule of periodic message transfers.
Reference [9] used pipeline technology with non-blocking communications based on message passing interface (MPI) to reduce waiting times in data synchronization among nodes.
Node-level real-time guarantee is achieved by high-resolution timer, interrupt/event priorities, resource scheduling, non-blocking communication, etc. Message-level real-time guarantee is achieved based on the network protocols, such as RT-NET, which provide real-time support.
First, SBA is a semi-blocking algorithm whereas other algorithms are either non-blocking or blocking algorithms.
We proposed a scheme called MSG, which provides users with a set of essential message-passing programming interfaces adopted from MPI and MCAPI, including blocking and non-blocking point-to-point communications, one-sided communications, and collective operations.
For example, with 1,000 slaves and 1 million information items on outages to be collected directly to a single point, while we adopt non-blocking sending and non-blocking receiving of messages to achieve computing and message passing simultaneously, the cost of maintaining sockets among the processors and handling random values replacing cost over master is considerable.
The major topics investigated include: blocking versus non-blocking communication, TCP/IP configuration options, multi-rail connections.
