|author||Steve Capper <email@example.com>||2013-06-10 16:58:27 +0100|
|committer||Steve Capper <firstname.lastname@example.org>||2013-06-11 08:02:06 +0100|
Linaro: Server config.
This is a set of options for a server configuration. Hopefully this will provoke some discussion on various sane defaults as some values aren't obvious to me! I merged these options as follows: ARCH=arm ./scripts/kconfig/merge_config.sh linaro/configs/linaro-base.conf \ linaro/configs/ubuntu-minimal.conf linaro/configs/arndale.conf \ linaro/configs/server.conf linaro/configs/kvm-host.conf \ linaro/configs/kvm-guest.conf Signed-off-by: Steve Capper <email@example.com>
1 files changed, 38 insertions, 0 deletions
diff --git a/linaro/configs/server.conf b/linaro/configs/server.conf
new file mode 100644
@@ -0,0 +1,38 @@
+# A set of config options for server workloads.
+# I merged this together with the Arndale config via:
+# ARCH=arm ./scripts/kconfig/merge_config.sh linaro/configs/linaro-base.conf \
+# linaro/configs/ubuntu-minimal.conf linaro/configs/arndale.conf \
+# linaro/configs/server.conf linaro/configs/kvm-host.conf \
+# A network bridge makes guest VM networking easier
+# Huge Pages can reduce TLB thrashing for certain workloads.
+# Transparent huge pages enabled but set to "madvise"
+# This means applications must ask for them with madvise syscall.
+# We do this as THPs can slow down short term tasks (by copying memory to huge
+# pages). For servers with long running tasks this can safely be enabeled.
+# One can enable THP for all cases via:
+# echo always > /sys/kernel/mm/transparent_hugepage/enabled
+# A point for discussion, with 3G of user space one can mmap larger files.
+# With 3G kernel space, more file caching can take place...
+# Set at 3G user space for the moment.
+# Save PTEs to high mem. This will make faulting new pages slightly more
+# expensive but gives us a little more normal memory headroom in the kernel
+# potentially speeding up other things.
+# Switch over to the SLUB allocator. It is likely to replace SLAB soon, is
+# a little simpler and faster, and is the default slab allocator in the
+# kernel since about 2.6.22 anyway...