System freezes on Blade 100/150 - possible due to file buffers
Connor McLaughlan <[email protected]> Mon, 4 Sep 2023 17:41:22 +0200
| Newsgroups | gmane.os.netbsd.ports.sparc64 |
|---|---|
| Message-ID | <CABo1Du7jV5u48tGoJp2LMkKi85jaT6ArbMKiUHrX6sC=JjeTWA@mail.gmail.com> |
Hello All, for a while now i am experiencing system freezes on a Blade 100 and a Blade 150 with NetBSDs from 9.0-9.3, seemingly related to files transfered to harddisk. If i don't copy anything to the harddisk from usb or network, the systems stay up forever. I tried to build all kinds of different kernels, falling back to generic controller drivers or disabling dma and udma. Nothing helped so far. I don't get any error, the systems just freeze. I suspect an issue with the buffer cache: I am experiencing the freezes also when i run the systems from a NetBSD install CD. However i found out, that if i keep the file buffer (as given by top - xxM File) below 800mb the systems will stay up and running. I only found a single method to evict the buffer caches - by unmounting and remounting the source and target drives, which seems to only be possible using a boot CD, as i found no option to remount a drive on a running system (is this possible in NetBSD, like in linux: mount -o remount,rw ...?) Also i tried to modify a few of the vm.* subtree values like vm.bufcache and vm.bufmem_hiwater via sysctl to see if this has any effect on the buffer cache evicting files or not growing too large, without success. Is there a command to release all the buffer caches or a possibility to disable caching? (I commented out both buffer strategies BUFQ_READPRIO and BUFQ_PRIOSCAN and also tried to set bufcache to 0 and built a new kernel, which did not disable the file buffering. Does anyone have any ideas about this behavior or experienced it also? Regards, Connor