Network

This section discusses network kernel parameters, including those for the following:

• Net core

• ICMP

• NEIGHBOR

• CONFIG

• Routing

• TCP

• IP fragmentation

Net Core

Sysctl Prefix: net.core

/proc Prefix: /proc/sys/net/core

Parameter	Description
hot_list_length	Maximum number of skb-heads to be cached
optmem_max	Maximum ancillary buffer size per socket
message_burst	Limits warning messages
message_cost	Limits warning messages
netdev_max_backlog	Maximum number of packets queued
rmem_default	Socket receive buffer size
wmem_default	Socket send buffer size
rmem_max	Maximum receive socket buffer size
wmem_max	Maximum send socket buffer size

The message_cost and message_burst parameters are used to limit the warning messages written to the kernel log. The default setting limits warning messages to one every 5 seconds. The higher the error_cost is, the fewer messages are written. The error_bust parameter controls when messages are dropped.

ICMP

Sysctl Prefix: net.ipv4

/proc Prefix: /proc/sys/net/ipv4

Parameter	Description
icmp_ratemask	Mask value for ICMP types
icmp_ratelimit	Rate for sending ICMP packets
icmp_ignore_bogus_error_responses	Warning messages from router
icmp_echo_ignore_broadcasts	ICMP echo requests to broadcast addresses
icmp_echo_ignore_all	ICMP echo requests
igmp_max_membership	Maximum number of multicast groups

The icmp_ratemask parameter contains ICMP types for which rates are limited.

The icmp_ratelimit parameter limits the maximum rates for sending ICMP packets whose type matches the icmp_ratemask parameter to specific targets. A value of 0 disables any limit. The value is in jiffies.

If icmp_ignore_bogus_error_responses is enabled, the kernel does not issue any log warnings when routers send bogus responses to broadcast frames.

If the icmp_echo_ignore_all parameter is enabled, the kernel ignores all the ICMP echo requests sent to it. If icmp_echo_ignore_broadcasts is enabled, the kernel ignores all the ICMP echo requests sent to broadcast/multicast addresses.

The igmp_max_membership parameter is the maximum number of multicast groups we can subscribe to.

NEIGHBOR

Sysctl Prefix: net.ipv4.neigh.interface

/proc Prefix: /proc/sys/net/neigh/interface/

Parameter	Description
locktime	Time before replacing an old ARP entry
proxy_delay	Maximum time before answering an ARP request
anycast_delay	Maximum delay for neighbor solicitation messages
proxy_qlen	Maximum queue length of delayed proxy ARP
unres_qlen	Maximum queue length for pending ARP request
gc_stale_time	Time before a check for stale ARP entries
delay_first_probe_time	Delay for first probe if neighbor is reachable
base_reachable_time	Validity period of an entry once a neighbor has been found. Default is 30 seconds.
retrans_time	Time between neighbor solicitation retransmits
app_solicit	Number of requests to send to user-level ARP daemon
ucast_solicit	Maximum number of retries for unicast solicitation
mcast_solicit	Maximum number of retries for multicast solicitation

CONFIG

Sysctl Prefix: net.ipv4.conf.interface

/proc Prefix: /proc/sys/net/conf/interface/

Parameter	Description
proxy_arp	Responds to ARP requests if enabled
rp_filter	Source validation is required if enabled
bootp_relay	Accepts packets with 0.x.y.z addresses
accept_source_route	Accepts set of IP addresses to visit on way
send_redirects	Sends ICMP redirects
shared_media	Different subnet communication
secure_redirects	Packet resend using secure interface
accept_redirects	Packet resend on same interface on an ICMP redirect
mc_forwarding	Multicast forwarding
forwarding	IP forwarding

The secure_redirects parameter accepts ICMP redirect messages only for a gateway listed in the default gateway list. The send_redirects parameter determines whether such redirects are sent.

Turning on the accept_source_route parameter gives a packet a list of IP addresses it should go through on its way.

The bootp_relay parameter accepts packets with source addresses of 0.x.y.z with destinations not to this host as local ones. The BOOTP daemon is expected to catch and forward such packets.

The mc_forwarding parameter enables or disables multicast forwarding on this interface.

The shared_media parameter allows the kernel to assume that different subnets on this device can communicate directly.

Routing

Sysctl Prefix: net.ipv4.route

/proc Prefix: /proc/sys/net/route/

Parameter	Description
error_burst	Time before logging a warning message
error_cost	Cost factor for logging a warning message
max_size	Maximum size of routing cache
max_delay	Maximum delay in seconds for flushing routing cache
min_delay	Minimum delay in seconds for flushing routing cache

The error_cost and error_bust parameters limit the warning messages written to the kernel log. The default setting limits warning messages to one every 5 seconds. The higher error_cost is, the fewer messages are written. The error_bust parameter controls when messages are dropped.

TCP

Sysctl Prefix: net.ipv4

/proc Prefix: /proc/sys/net/ipv4/

Parameter	Description
tcp_app_win	Maximum window for application buffer
tcp_reordering	Maximum reordering of packets in a TCP stream
tcp_max_syn_backlog	Length of per-socket backlog queue
tcp_rfc1337	Sets TCP stack confirming to RFC1337
tcp_stdurg	Enables strict RFC793 interpretation of the TCP urgent pointer field
tcp_fin_timeout	Time in seconds it takes to receive a final FIN before the socket closes
tcp_tw_reuse	Reuse of timewait sockets
tcp_max_orphans	Maximum number of orphaned connections
tcp_max_tw_buckets	Maximum number of timewait sockets
tcp_window_scaling	Enables window scaling as defined in RFC1323
tcp_timestamps	Enables timestamps as defined in RFC1323
tcp_rmem	Receive buffers for TCP socket
tcp_wmem	Send buffers for TCP socket
tcp_mem	Memory pages for TCP usage

The tcp_wmem parameter has three fields. The first field, min, specifies the amount of memory reserved for send buffers for a TCP socket. The second field is the default amount of memory allowed for send buffers for a TCP socket. The last field, max, is the maximum amount of memory allowed for automatically selected send buffers for a TCP socket.

The tcp_rmem parameter has three fields. The first field, min, is the minimum size of receive buffer used by TCP sockets. The second is the default amount of memory allowed for receive buffers used by TCP sockets. The last field, max, is the maximum size of the receive buffer allowed for automatically selected receive buffers for TCP sockets.

The tcp_mem parameter has three fields. The first field, low, specifies the low limit on the number of pages below which TCP is not bothered about its usage of memory pages. The second field is high; when the amount of memory allocated by TCP exceeds this, TCP takes corrective actions until memory consumption falls below the low value. The last field, max, is the maximum number of pages allowed for queuing by all TCP sockets.

Parameter	Description
tcp_sack	Enables sack to acknowledge when a packet is received
tcp_fack	Enables forward acknowledgment
tcp_dsack	Enables extension to sack for duplicate packet detection
ecn	Enables explicit congestion notification

Parameter	Description
tcp_keepalive_intvl	Frequency of keepalive probes
tcp_keepalive_probes	Number of keepalive probes
tcp_keepalive_time	Keepalive message interval
tcp_syn_retries	Number of initial SYNs for outgoing connection
tcp_retries1	Number of retries for incoming connections
tcp_retries2	Number of retries before terminating

The tcp_keepalive_probes parameter is the number of times the probes are sent before killing the nonresponding connections.

The tcp_syn_retries parameter is the number of times initial SYNs for a TCP connection are transmitted. This timeout is for outgoing connections. For incoming connections, the number of retransmits is defined by the tcp_retries1 parameter.

IP Fragmentation

Sysctl Prefix: net.ipv4

/proc Prefix: /proc/sys/net/ipv4

Parameter	Description
ip_nonlocal_bind	Binds to nonlocal IP address
ip_no_pmtu_disc	Disables MTU path discovery
ip_autoconfig	How host got IP configuration
ip_default_ttl	Maximum number of hops a packet may travel
ip_forward	Forward packets between interfaces
ipfrag_time	Time in seconds to keep IP fragment in memory
ip_dynaddr	Enables dynamic socket address rewriting on interface address change
ipfrag_low_thresh	Low-memory watermark for fragment handler
ipfrag_high_thresh	Maximum amount of memory used to reassemble IP fragments
ip_local_port_range	Range of ports used by TCP and UDP to choose the local port