...COMM_WORLD
It has been pointed out that if multiple MPI threads are allowed in the same Java VM, the default communicator cannot be obtained from a static variable. The final version of the API may change this convention.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...data
A better protocol would be to eagerly send data for short messages in the header, assuming some fixed-size buffer is preallocated at the receiving end. The two-message protocol would be reserved for long messages. This marginally complicates the implementation but does not essentially change the rest of the discussion or the benchmark results presented below, since the latter concentrate on the asymptotic case. We are grateful to one of the anonymous referees for raising this point.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...elements
We note that there some debate about whether the Java model of multidimensional arrays is the most appropriate one for high performance computing. There are various proposals for for optimized HPC array class libraries [15]. See section 6 for some further discussion.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...T[N][N]
Our timing model assumed the values of these parameters is independent of the element type. This is only approximately true, and the values quoted in the table and used in the plotted curves are averages. Separately measured values for byte arrays were smaller than these averages, and for int and float arrays they were larger.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...elements
Defined to be arrays of objects, each element being an array of primitive type of the same type and length.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...process
It can also introduce some unexpected behaviour. Our version subtly alters the semantics of serialization, because it does not detect of aliasing of rows (either with other rows of the same two-dimensional array, or with one-dimensional primitive arrays elsewhere in the stream). Hence the reconstructed object graph at the receiving end will not reproduce such aliasing. Whether this is a serious problem is unclear.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

Bryan Carpenter
Thu Nov 4 13:48:00 EST 1999