4mb PAGES for mbuf clusters

km b kmb810 at gmail.com
Sat Jul 14 23:38:04 PDT 2007

On 7/15/07, Matthew Dillon <dillon at apollo.backplane.com> wrote:
    Alan announced it a few months ago. I'm trying to find the reference.
    I'll bet you can search the freebsd mail archives to find it.
Thanks. Now I know who announced it. Maybe it'll make searching easier.

    Again, though, the issue is not so much implementing the large
    pages but implementing mechanisms that prevent fragmentation from
    making it impossible to construct physically contiguous pages that
    are that large.
To avoid memory fragmentation what I am thinking is that at system
startup reserve two memory pools for mbufs and use just one of them.
When it becomes fragmented (upto 80% or user configurable), switch to
the next one and recover the fragmented memory. However, this may not
be a general solution but at least it may work for specific workloads.
Something is wrong up on cloud # 9!
Something is wrong up on cloud # 9!

More information about the Kernel mailing list