Konrad Neuwirth <konrad%mailathome.or.at@localhost> writes: >>> No. The only result that doesn't produce an Invalid argument is … 16384. >>> >>> What can I do? >>> >>> This is a NetBSD/i386 machine with 2GB physical RAM. >> >> Probably you are hitting hard limits in the code. I would look through >> the kernel source. Beware that just increasing the NMBCLUSTERS without >> also increasing kernel virtual address space is likely to lead to KVA >> exhaustion, which is not pretty, and manifests itself as a machine which >> is not quite crashed but won't do much. > > So would this problem be remedied if I switched the system from i386 > to amd64? Would the limits be more to my tastes, or do they behave > similarly under the same memory conditions? I suspect amd64 will be better, because there's no longer a good reason to keep KVA so small. But I don't really know that. > I still do not quite get it to fit in my head that we didn't have the > problem with NetBSD 5; we could just set the NMBCLUSTERS to something > quite bigger there and be happy. It's relatively easy to read the diff between the files that have the limit calculations. It's entirely possible that new checking code to constrain the limits was added, which increases safety but makes it more likely you will run out. There's unfortunately no way to understand without understanding the details.
Description: PGP signature