4 General Caveats

The CM-5 and Alewife implementations of CRL only support `single-threaded' applications (applications with a single thread/process per processor). The TCP/UNIX version allows a single thread for each PVM process (but multiple PVM processes may reside on the same physical machine).

For the most part, this implementation of CRL fails catastrophically if a problem is encountered instead of doing something reasonable like returning an informative error code.

After calling crl_init, applications should use safe_malloc and safe_free instead of malloc and free when allocating or deallocating memory. (See Section 3.1.6 for details.)

As distributed, the source code for CRL 1.0 will not work on systems with larger than 256 processors. (See the #define for MaxNodesBits in crl_int.h.)

See the ``Platform-Specific Issues'' subsections in each of the following sections for other caveats specific to a particular platform.