lkaplan@BBN.COM (Larry Kaplan) (02/20/91)
In article <13202@hubcap.clemson.edu> tve%allspice.Berkeley.EDU@ucbvax.Berkeley.EDU (Thorsten von Eicken) writes: > > I just read in an old ('85) paper that the RP3 supports interleaving >of memory across processors and I wonder how this interacts with >memory management (e.g. malloc or kernel-malloc). > .... > My question now is how do you do memory allocation in this >2-dimensional playground (Y-axis:local address, X-axis:processor)? >Are there any papers published on this??? The RP3 interleaving scheme, and others including BBN's interleaving scheme, allow the system to be configured so that only some of the memory is interleaved. This leads to having N+1 separate heaps for mallocs, one for each memory (node) and one for interleaved memory. Our approach has been to provide an extended malloc package for both the user and the kernel which allows the caller to ask for memory local to a specific node or interleaved memory. Of course, this places a lot of burden on the programmer but then he usually knows best. Some work has been done by various groups, e.g. Rochester and Duke, in trying to automatically handle the non-uniform memory management issues. These don't generally address the use of interleaved memory though. Hope this answers your question at least a little. _______________________________________________________________________________ ____ \ / ____ Laurence S. Kaplan | \ 0 / | BBN Advanced Computers lkaplan@bbn.com \____|||____/ 10 Fawcett St. (617) 873-2431 /__/ | \__\ Cambridge, MA 02238