ZFS kmem_map too small.

Oliver Fromme olli at lurza.secnetix.de
Mon Nov 12 09:35:03 PST 2007

[Excessive Cc list cut]

Ali Mashtizadeh wrote:
 > For ZFS KVA_PAGES=512 seems to run very stable so it's unfortunate that we
 > have to give up our VA space in the userland. I would personally like to see
 > a move to a 4-4 Memory model or at least have it as a compile time option.
 > Does anyone know if that's been attempted on FreeBSD?

Sorry for the late reply.  I haven't seen an answer to your
question, so I try to give one.

Using a 4-4 memory model (in 32 bit mode) means that kernel
and user pages could never be active at the same time, i.e.
they would have to be swapped for _every_ context switch,
including syscalls and interrupts.  That would be extremely
expensive.  And there would still be an address space limit
at 4 GB, so the problem would only be pushed a little farther
away, but it wouldn't be solved.

A much better solution (or rather: the _real_ solution) is to
use a 64bit OS (such as FreeBSD/amd64), where there's no need
to worry about address space splitting at all.  That's also
precisely the reason why amd64 is recommended over i386 when
using ZFS on FreeBSD.

Practically all new hardware is 64bit-capable, so that's the
way to go.


Oliver Fromme, secnetix GmbH & Co. KG, Marktplatz 29, 85567 Grafing b. M.
Handelsregister: Registergericht Muenchen, HRA 74606,  Geschäftsfuehrung:
secnetix Verwaltungsgesellsch. mbH, Handelsregister: Registergericht Mün-
chen, HRB 125758,  Geschäftsführer: Maik Bachmann, Olaf Erb, Ralf Gebhart

FreeBSD-Dienstleistungen, -Produkte und mehr:  http://www.secnetix.de/bsd

"And believe me, as a C++ programmer, I don't hesitate to question
the decisions of language designers.  After a decent amount of C++
exposure, Python's flaws seem ridiculously small." -- Ville Vainio

More information about the freebsd-current mailing list