On Tue, 24 Nov 2009, Mindaugas Kavaliauskas wrote: Hi,
> Thank You for test results. I like the simplicity of DLMT, and I > think this implementation is better than trying to adapt more > complex PT*, NED* or other ALLOCs. They are also using thread local mspaces. The main difference is caused by the fact that they are designed as more general solutions. > 16 separate mspaces is enough for > most applications. Only on some heavy loaded servers we may need > more active threads (minority of us have such number of CPU). This > number 16 can be increased for a specific applications. Perhaps the > only drawback of current implementation is that unused mspaces are > destroyed only on application exit, but it this can be improved I do not think it's problem. DLMALLOC release unused memory blocks to the system so the overhead is minimal if all allocated memory in given mspace is freed. > using some logic dependent on GC activation or thread destruction. See above and do not forget that memory allocated by one thread can be accessible by other threads so you cannot free mspaces when thread is destroyed. Anyhow if you want to force freeing unused mspaces which do not have any active memory blocks then it can be done quite easy by adding small function which will scan all allocated mspaces and free the ones which do not have any memory "in use". This function can be activated by GC and it can be extended to make also memory trimming for all existing mspaces (dlmalloc_trim()). best regards, Przemek _______________________________________________ Harbour mailing list (attachment size limit: 40KB) Harbour@harbour-project.org http://lists.harbour-project.org/mailman/listinfo/harbour
