I recently encountered extremly slow compilation and linking times with g++ 4.1 and ld 2.16. Suprisingly, on a different (and slower) machine with g++/gcc 3.4 and ld 2.15 compiling and linking was much faster (easily with a factor of 10 to 20). Since it slowed down the development, it was really annoying. So I was looking around on the web to find some fixes or help. I also tried newer versions of ld and g++, but it didn’t really change anything. Interestingly the CPU usage during linking was down at 2-3%. So finally I found the bottleneck with our NFS which caused a slow-down when using the option -m32 during linking (I’m on a 64 bit machine compiling in 32 bit, so I need this option). I found another bottleneck with the assembler (as) also caused by using NFS. Compiling to a file in the network (i.e., not local) slows down the assembler quite a bit. So as a workaround one should try to compile intermediate object (.o) files as well as to link object files to executables to some place in the local file system (i.e., the file given by the option -o should be somewhere locally).

I thought I should just put this note online, it might save some of our time in case you’re experiencing similar problems :).