On Thu, 25 Dec 2003, [gb2312] Xin LI/ÀîöÎ wrote: > First I would say "thank you" to those who helped me to find the cause of > the problem. Now I have confirmed that the problem goes away after a kernel > configuration tuning. > > To trigger the problem, the box must have more than 2GB RAM and the > following configuration: > > options MAXDSIZ=(1024UL*1024*1024) > options MAXSSIZ=(128UL*1024*1024) > options DFLDSIZ=(1024UL*1024*1024) > > A default, say, GENERIC kernel would not trigger panic's during my test. > > The problem exists on 4-STABLE and 5-CURRENT, however, by increasing kernel > virtual address space by adding a line like the following: > > options KVA_PAGES=512 > > The panics goes away. > I'd really love getting rid of most hardcoded memory allocations in the fbsd kernel, with the main bugger being NMBCLUSTERS, which one always forgets about untill it's too late :(. In solaris auto-tuning has been a top priority for a long while, and propper dynamic tuning should get it right ( or even better ) in 99% of the cases. People with other needs should be able to build a custom kernel. Maybe a nice goal for 6.0 :) I do have to note that fbsd5 has been getting better and better, on sparc64 it beats out linux and netbsd for my purposes. Good work. -- Sten Spans There is a crack in everything that's how the light gets in.Received on Thu Dec 25 2003 - 09:24:21 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:35 UTC