# created by Stefan Kaiser
# date: 2020-02-05
# related SAP note: 2382421
# to reload sysctl configuration you can use: #sysctl --system
# This sets the max OS receive buffer size for all types of connections.
net.core.rmem_max = 64000000
# This parameter limits the size of the accept backlog of a listening socket.
# The Linux default of 128 is not sufficient so you need to set the parameter to at least 4096 in order that the HANA system can use higher values.
# There is an interdependency between this parameter and HANA configuration parameter tcp_backlog. If net.core.somaxconn is set to a lower value than tcp_backlog, tcp_backlog will be silently truncated to the value set for net.core.somaxconn. Therefore, you need to ensure that net.core.somaxconn is always set to a value equal to or greater than tcp_backlog.
net.core.somaxconn = 4096
# These settings define the maximum socket send and receive buffer size.
# To ensure complete functionality it must be ensured that the wmem_max and rmem_max values are at least the same as the respective maximum value of the parameters net.ipv4.tcp_wmem and net.ipv4.tcp_rmem.
net.core.wmem_max = 64000000
# This is the size of the SYN backlog.
# To prevent the kernel from using SYN cookies in a situation where lots of connection requests are sent in a short timeframe and to prevent a corresponding warning about a potential SYN flooding attack in the system log, the size of the SYN backlog should be set to a reasonably high value.
net.ipv4.tcp_max_syn_backlog = 8192
# These parameters specify the minimum, default and maximum size of the TCP send and receive buffer.
net.ipv4.tcp_rmem = 10000000 10000000 10000000
net.ipv4.tcp_wmem = 10000000 10000000 10000000
# This setting disables the need to scale-up incrementally the TCP window size for TCP connections which were idle for some time. Using this parameter it is ensured that the maximum speed is used from beginning also for previously idle TCP connections.
net.ipv4.tcp_slow_start_after_idle = 0
# The default value for this parameter is 5, which translates to a timeout of about 24 seconds.
# If the system is under load, a timeout of 24 seconds can be too short and lead to avoidable errors.
# It also prevents processes to set a longer timeout. The recommended value is 8, which translates into a timeout of 190 seconds
net.ipv4.tcp_syn_retries = 8
# This setting adds the timestamp field to the TCP header.
# It should already be active on most systems and is a prerequisite for net.ipv4.tcp_tw_reuse and net.ipv4.tcp_tw_recycle.
net.ipv4.tcp_timestamps = 1
# This setting reduces the time a connection spends in the TIME_WAIT state.
# One precondition for it to take effect is that TCP timestamps are enabled, i.e. net.ipv4.tcp_timestamps = 1, which is the default on most modern systems.
net.ipv4.tcp_tw_recycle = 1
# This setting allows HANA to reuse a client port immediately after the connection has been closed, even though the connection is still in TIME_WAIT state. A precondition for it to take effect is that TCP timestamps are enabled, i.e. net.ipv4.tcp_timestamps = 1, which is the default on most modern systems.
net.ipv4.tcp_tw_reuse = 1
# This setting enables the TCP window scaling.
# On most systems it already should be active. Moreover, it is a prerequisite for net.ipv4.tcp_wmem and net.ipv4.tcp_rmem.
net.ipv4.tcp_window_scaling = 1
# As HANA uses a considerable number of connections for the internal communication, it makes sense to have as many client ports available as possible for this purpose.
# At the same time, you need to ensure that you explicitly exclude the ports used by processes and applications which bind to specific ports by adjusting parameter net.ipv4.ip_local_reserved_ports accordingly.
net.ipv4.ip_local_port_range = 9000 64999
# SAP Host Agent takes care of adjusting this parameter and setting it manually is neither recommended nor required
#net.ipv4.ip_local_reserved_ports =
# SAP HANA is a NUMA (non-uniform memory access) aware database. Thus it does not rely on the Linux kernel's features to optimize NUMA usage automatically. Depending on the workload, it can be beneficial to turn off automatical NUMA balancing.
kernel.numa_balancing = 0
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
User | Count |
---|---|
10 | |
9 | |
5 | |
4 | |
4 | |
3 | |
3 | |
3 | |
3 | |
3 |