# ktune sysctl settings for EL 5 servers

# 256 KB default performs well experimentally, and is often recommended by ISVs.
net.core.rmem_default = 262144
net.core.wmem_default = 262144

# When opening a high-bandwidth connection while the receiving end is under
# memory pressure, disk I/O may be necessary to free memory for the socket,
# making disk latency the effective latency for the bandwidth-delay product
# initially.  For 10 Gb ethernet and SCSI, the BDP is about 5 MB.  Allow 8 MB
# to account for overhead, to ensure that new sockets can saturate the medium
# quickly.
net.core.rmem_max = 8388608
net.core.wmem_max = 8388608

# Allow a deep backlog for 10 Gb and bonded Gb ethernet connections
net.core.netdev_max_backlog = 10000

# Always have one page available, plus an extra for overhead, to ensure TCP NFS
# pageout doesn't stall under memory pressure.  Default to max unscaled window,
# plus overhead for rmem, since most LAN sockets won't need to scale.
net.ipv4.tcp_rmem = 8192 87380 8388608
net.ipv4.tcp_wmem = 8192 65536 8388608

# Always have enough memory available on a UDP socket for an 8k NFS request,
# plus overhead, to prevent NFS stalling under memory pressure.  16k is still
# low enough that memory fragmentation is unlikely to cause problems.
net.ipv4.udp_rmem_min = 16384
net.ipv4.udp_wmem_min = 16384

# Ensure there's enough memory to actually allocate those massive buffers to a
# socket.
net.ipv4.tcp_mem = 8388608 12582912 16777216
net.ipv4.udp_mem = 8388608 12582912 16777216

# Filesystem I/O is usually much more efficient than swapping, so try to keep
# swapping low.  It's usually safe to go even lower than this on systems with
# server-grade storage.
vm.swappiness = 30

# If a workload mostly uses anonymous memory and it hits this limit, the entire
# working set is buffered for I/O, and any more write buffering would require
# swapping, so it's time to throttle writes until I/O can catch up.  Workloads
# that mostly use file mappings may be able to use even higher values.
vm.dirty_ratio = 50

# Ensure there's always some easily-dropped pagecache if the system is under
# memory pressure from cached files, since it's much faster to page back in than
# swap.
# Doesn't exist for 2.6.30++ anymore
# vm.pagecache = 90