Peter Wemm writes:
> I believe we have bigger problems in the MI code.. we allocate the
> vm_page_t array to cover a linear range from the lowest entry in
> phys_avail to the highest entry address. This means that if we have
> machines that have 2G of ram at address 0 and another 2G at the 16G mark,
> then that means that the MI vm code allocates enough vm_page_t's in a
> linear array to to cover 18G of physical space. Needless to say, this
That's exactly my situation. The alpha I'm porting to puts up to 16GB
of RAM per-cpu. So a 2 CPU system with nGB/cpu has a phys_avail that
looks like: {0,nGG,16GB,(16+n)GB}. Eg, a huge (16 - n)GB hole between
each CPU. This could eat a lot of memory on a 64-cpu box.
BTW, I think I may have seen some autotuning issues with an 8GB box.
nbufs gets set to something absurd and an apparent infinate loop is
entered in vfs_bio.c, for example. Have you ia64 guys run with 8GB
yet?
I'm not going to worry about it too much just yet, as I need to set
MAXMEM to 1GB anyway because alpha's busdma support is incomplete. (as
is the busdma support for if_dc, which is what I'm booting from..).
Drew
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe cvs-src" in the body of the message