Skip to content
GitLab
Menu
Projects
Groups
Snippets
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
Menu
Open sidebar
emulab
emulab-devel
Commits
d9b96847
Commit
d9b96847
authored
Jun 23, 2017
by
Mike Hibler
Browse files
Kernel mbuf sizing no longer needed as of FreeBSD 10.x.
parent
46c27d18
Changes
3
Hide whitespace changes
Inline
Side-by-side
clientside/tmcc/freebsd10/loader.conf
View file @
d9b96847
...
...
@@ -4,10 +4,6 @@ beastie_disable="YES"
# Shorten this to allow faster boots
autoboot_delay
=
"5"
## for wireless nodes
#if_ath_load="YES"
#hw.bge.allow_asf=1
#
# If you are using an extra, regular (i.e., visible to the OS) Ethernet port
# for your management interface, then you should disable it so that FreeBSD
...
...
@@ -26,35 +22,13 @@ hw.memtest.tests=0
hw
.
mfi
.
mrsas_enable
=
1
#
# Some drivers, most notably the Intel 10Gb "ixgbe" driver and newer 1Gb
# drivers, are very liberal with their mbuf allocation for RX buffers.
# The ixgbe driver in particular wants to allocate up to 2048 buffers
# per interface per CPU core (up to 8). On the Utah d820 nodes that is
# 2048 * 4 * 8 == 65536 mbufs.
#
# This can cause dhclient to fail when it fires up all interfaces at once
# and the kernel runs out of mbufs.
#
# At Utah we avoid this by not dhcp'ing on the "ix" interfaces (see
# /etc/rc.d/netif-emulab) because they are never the control net. But if
# you have a 10Gb control net, or some other interface that allocates
# excessive mbufs, you may need to try the following settings or something
# even larger. Note that the relationship between the four values is the
# same as what FreeBSD uses for the defaults; i.e., each value is half the
# preceding value.
#
# Someday, these values will get re-adjusted in the Emulab startup scripts
# to a more appropriate value based on the size of RAM and total number of
# interfaces.
#
# As of January 2016, we now have new nodes that have a total of 6
# problematic interfaces each along with 16 cores, so we had to bump this
# even higher!
# N.B. As of FreeBSD 10.1, you don't need to tweak these, the defaults
# are plenty big!
#
kern
.
ipc
.
nmbclusters
=
262144
kern
.
ipc
.
nmbjumbop
=
131072
kern
.
ipc
.
nmbjumbo16
=
65536
kern
.
ipc
.
nmbjumbo9
=
32768
#
kern.ipc.nmbclusters=262144
#
kern.ipc.nmbjumbop=131072
#
kern.ipc.nmbjumbo16=65536
#
kern.ipc.nmbjumbo9=32768
# force module path, our pxeboot can mess it up
module_path
=
"/boot/kernel;/boot/modules"
...
...
clientside/tmcc/freebsd11/loader.conf
View file @
d9b96847
...
...
@@ -22,35 +22,13 @@ hw.memtest.tests=0
hw
.
mfi
.
mrsas_enable
=
1
#
# Some drivers, most notably the Intel 10Gb "ixgbe" driver and newer 1Gb
# drivers, are very liberal with their mbuf allocation for RX buffers.
# The ixgbe driver in particular wants to allocate up to 2048 buffers
# per interface per CPU core (up to 8). On the Utah d820 nodes that is
# 2048 * 4 * 8 == 65536 mbufs.
# N.B. As of FreeBSD 10.1, you don't need to tweak these, the defaults
# are plenty big!
#
# This can cause dhclient to fail when it fires up all interfaces at once
# and the kernel runs out of mbufs.
#
# At Utah we avoid this by not dhcp'ing on the "ix" interfaces (see
# /etc/rc.d/netif-emulab) because they are never the control net. But if
# you have a 10Gb control net, or some other interface that allocates
# excessive mbufs, you may need to try the following settings or something
# even larger. Note that the relationship between the four values is the
# same as what FreeBSD uses for the defaults; i.e., each value is half the
# preceding value.
#
# Someday, these values will get re-adjusted in the Emulab startup scripts
# to a more appropriate value based on the size of RAM and total number of
# interfaces.
#
# As of January 2016, we now have new nodes that have a total of 6
# problematic interfaces each along with 16 cores, so we had to bump this
# even higher!
#
kern
.
ipc
.
nmbclusters
=
262144
kern
.
ipc
.
nmbjumbop
=
131072
kern
.
ipc
.
nmbjumbo16
=
65536
kern
.
ipc
.
nmbjumbo9
=
32768
#kern.ipc.nmbclusters=262144
#kern.ipc.nmbjumbop=131072
#kern.ipc.nmbjumbo16=65536
#kern.ipc.nmbjumbo9=32768
# force module path, our pxeboot can mess it up
module_path
=
"/boot/kernel;/boot/modules"
...
...
clientside/tmcc/freebsd12/loader.conf
View file @
d9b96847
...
...
@@ -22,35 +22,13 @@ hw.memtest.tests=0
hw
.
mfi
.
mrsas_enable
=
1
#
# Some drivers, most notably the Intel 10Gb "ixgbe" driver and newer 1Gb
# drivers, are very liberal with their mbuf allocation for RX buffers.
# The ixgbe driver in particular wants to allocate up to 2048 buffers
# per interface per CPU core (up to 8). On the Utah d820 nodes that is
# 2048 * 4 * 8 == 65536 mbufs.
# N.B. As of FreeBSD 10.1, you don't need to tweak these, the defaults
# are plenty big!
#
# This can cause dhclient to fail when it fires up all interfaces at once
# and the kernel runs out of mbufs.
#
# At Utah we avoid this by not dhcp'ing on the "ix" interfaces (see
# /etc/rc.d/netif-emulab) because they are never the control net. But if
# you have a 10Gb control net, or some other interface that allocates
# excessive mbufs, you may need to try the following settings or something
# even larger. Note that the relationship between the four values is the
# same as what FreeBSD uses for the defaults; i.e., each value is half the
# preceding value.
#
# Someday, these values will get re-adjusted in the Emulab startup scripts
# to a more appropriate value based on the size of RAM and total number of
# interfaces.
#
# As of January 2016, we now have new nodes that have a total of 6
# problematic interfaces each along with 16 cores, so we had to bump this
# even higher!
#
kern
.
ipc
.
nmbclusters
=
262144
kern
.
ipc
.
nmbjumbop
=
131072
kern
.
ipc
.
nmbjumbo16
=
65536
kern
.
ipc
.
nmbjumbo9
=
32768
#kern.ipc.nmbclusters=262144
#kern.ipc.nmbjumbop=131072
#kern.ipc.nmbjumbo16=65536
#kern.ipc.nmbjumbo9=32768
# force module path, our pxeboot can mess it up
module_path
=
"/boot/kernel;/boot/modules"
...
...
Write
Preview
Supports
Markdown
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment