Communication Time

Predicted and measured parallel eFficiencies for two-electron integral computation using dynamic distribution of shell pairs. The predicted efficiency for static distribution of shell pairs is included for comparison. Results were obtained for C4H10 with the cc-pVTZ basis set. 

---

INPUTS OUTPUTS COMMUNICATIONS TIME TASK DUTIES, ... 

Good scaling as a function of the number of processors will occur when the ratio of the communication time to time spent in the local portion of the code is small. We give a good rough estimate of this ratio as [40]... 

These refer to processes, procedures, and systems for communicating timely and accurate information to the public during crisis or emergency situations. 

The results obtained from the simulation of the integrated operation were utilized for determining the parallelism of the tasks. The parallelism optimization assumes the tasks duration and communication times as per those generated by the TIE 1.4 simulation (Table 5) and congestion time as 0.02e° The results obtained are presented in Table 7 and Figures 11 and 12. 

The Rabenseifner all-reduce algorithm is identical to the Rabenseifner all-to-one reduction, except for using an all-to-all broadcast (allgather) operation in place of gather, and the communication time for this algorithm can be modeled as... 

The network performance characteristics for a parallel computer may greatly influence the performance that can be obtained with a parallel application. The latency and bandwidth are among the most important performance characteristics because their values determine the communication overhead for a parallel program. Let us consider how to determine these parameters and how to use them in performance modeling. To model the communication time required for a parallel program, one first needs a model for the time required to send a message between two processes. For most purposes, this time can... 

Let us develop an expression for the execution time for a parallel program, assuming that computation is not overlapped with communication and that no process is ever idle. Each process, then, will always be engaged in either computation or communication, and the execution time can be expressed as a sum of the computation and the communication times... 

To determine values for the machine parameters a, p, and y, a series of test runs were performed using a Linux cluster. The value for y was estimated to be 4.3 ns by timing single-process matrix-vector multiplications for various matrix sizes. To model the communication time, the values of a and the sum ip + y are required these values were found to be a = 43 jxs and 2p + Y = 72 ns/word (using 8 byte words) by timing the all-reduce operation as a function of the number of processes for a number of problem sizes and fitting the data to a function of the form of Eq. 5.20. Using these values for the machine parameters, the performance model was used... 

We have used expressions involving the latency, a, and inverse bandwidth, /3, to model the communication time. An alternative model, the Hockney model, is sometimes used for the communication time in a parallel algorithm. The Hockney model expresses the time required to send a message between two processes in terms of the parameters Too and ni, which represent the asymptotic bandwidth and the message length for which half of the asymptotic bandwidth is attained, respectively. Metrics other than the speedup and efficiency are used in parallel computing. One such metric is the Karp-Flatt metric, also referred to as the experimentally determined serial fraction. This metric is intended to be used in addition to the speedup and efficiency, and it is easily computed. The Karp-Flatt metric can provide information on parallel performance characteristics that caimot be obtained from the speedup and efficiency, for instance, whether degrading parallel performance is caused by incomplete parallelization or by other factors such as load imbalance and communication overhead. ... 

Interspersion of computation and communication time on a process in a fine-grained and a coarse-grained parallel algorithm. 

Using the earlier expressions for the computation time and the communication time, we can derive the efficiency for the algorithm. First, note that the efficiency can be expressed as follows... 

Log-log plot of predicted total execution times (Pltotal and P2totai) and communication times (Plcomm and P2comm) on a Linux cluster for the uracil dimer using the cc-pVDZ basis set. Ideal scaling corresponds to a straight line of negative unit slope. 

The measured total execution times and communication times for PI and P2 on the employed Linux cluster" are illustrated in Figure 9.7. For larger... 

---