kern/108375: net.inet.raw.maxdgram shouldn't be less than MTU
Bruce M. Simpson
bms at FreeBSD.org
Tue Mar 20 12:21:32 UTC 2007
Vladimir Ivanov wrote:
> Increasing default buffer size from 8k upto 16k seems to be more
> essential way.
> Isn't it?
> I've never seen jumbo packets more than 9k still.
Sorry the penny dropped after I sent that reply!
In -CURRENT, the default socket send buffer size for UDP is 9216 bytes.
The default send and receive socket buffer size for raw sockets is 8192
bytes. A 9000 byte mtu jumbogram is going to be fragmented with this.
It would seem reasonable to increase the raw size. An 8192 byte buffer
is going to fit into more than 2 i386 pages anyway when using clusters,
so this is no big deal and seems like the right change.
More information about the freebsd-bugs