AdminSysGNULinux

Outils pour utilisateurs

Outils du site

Piste :
tech:reseau_linux_pile_tcp_ip

Différences

Ci-dessous, les différences entre deux révisions de la page.

Lien vers cette vue comparative

Prochaine révision		Révision précédente
tech:reseau_linux_pile_tcp_ip [2025/03/24 15:06] – créée - modification externe 127.0.0.1tech:reseau_linux_pile_tcp_ip [2026/06/30 21:54] (Version actuelle) – Jean-Baptiste
Ligne 1: Ligne 1:
 +<!DOCTYPE markdown>
 +{{tag>Brouillon Réseau Linux Kernel TCP CA}}
 +
 +
 +# Réseau Linux pile TCP/IP
 +
 +Voir aussi :
 +* MPTCP, SCTP, DCCP
 +
 +
 +Voir :
 +* https://docs.kernel.org/networking/ip-sysctl.html
 +* https://www.inetdoc.net/guides/lartc/lartc.kernel.obscure.html
 +* https://man7.org/linux/man-pages/man7/tcp.7.html
 +* https://frsag.frsag.narkive.com/IYrhTt32/incidence-des-tcp-timestamps-sur-les-connexions
 +* hping2
 +* https://github.com/zersh01/linux-tcp-queues-and-metrics/blob/main/README_EN.md
 +
 +~~~bash
 +man 7 tcp
 +~~~
 +
 +
 +## Contrack
 +
 +Voir :
 +* /proc/net/nf_conntrack
 +* /proc/sys/net/nf_conntrack_max
 +
 +~~~bash
 +apt-get install conntrack
 +~~~
 +
 +Flush
 +~~~bash
 +conntrack -F
 +~~~
 +
 +
 +
 +### /proc/sys/net/ipv4/tcp_syn_retries
 +
 +
 +~~~
 +$ sysctl net.ipv4.tcp_syn_retries
 +net.ipv4.tcp_syn_retries = 6
 +~~~
 +
 +Effectively, this takes 1+2+4+8+16+32+64=127s before the connection finally aborts.
 +
 +
 +### /proc/sys/net/ipv4/tcp_synack_retries
 +
 +
 +
 +### /proc/sys/net/ipv4/tcp_retries2
 +
 +Voir : 
 +* https://stackoverflow.com/questions/5227520/how-many-times-will-tcp-retransmit
 +* https://pracucci.com/linux-tcp-rto-min-max-and-tcp-retries2.html
 +
 +Voir aussi :
 +* /proc/sys/net/ipv4/tcp_retries
 +* /proc/sys/net/ipv4/tcp_syn_retries
 +* /proc/sys/net/ipv4/tcp_synack_retries
 +
 +#### Cluster
 +
 +In a High Availability (HA) situation consider decreasing the setting to 3.
 +
 +RFC 1122 recommends at least 100 seconds for the timeout, which corresponds to a value of at least 8.
 +Oracle suggest a value of 3 for a RAC configuration.
 +
 +Source : https://access.redhat.com/solutions/726753
 +
 +#### Nb de retransmissions vs temps
 +
 +An experiment confirms that (on a recent Linux at least) the timeout is more like 13s with the suggested net.ipv4.tcp_retries2=5
 +
 +“Windows defaults to just 5 retransmissions which corresponds with a timeout of around 6 seconds.”
 +“Five retransmissions corresponds with a timeout of around six seconds.”
 +tcp_retries2=5 means timeout with first transmission plus 5 retransmissions: 12.6 seconds=(2^6 - 1) * 0.2.
 +tcp_retries2=15: 924.6 seconds=(2^10 - 1) * 0.2 + (16 - 10) * 120.
 +
 +
 +Source : https://github.com/elastic/elasticsearch/issues/102788
 +
 +Voir aussi : https://www.elastic.co/guide/en/elasticsearch/reference/current/system-config-tcpretries.html#_related_configuration
 +
 +
 +### F_RTO
 +
 +https://access.redhat.com/solutions/4978771
 +
 +
 +
 +## TCP keepalive 
 +
 +Voir :
 +* https://blog.cloudflare.com/when-tcp-sockets-refuse-to-die
 +  * Python Scripts https://github.com/cloudflare/cloudflare-blog/tree/master/2019-09-tcp-keepalives
 +
 +tcp_keepalive_time
 +https://www.veritas.com/support/en_US/article.100028680
 +### Configuring TCP/IP keepalive parameters for high availability clients (JDBC)
 +
 +  *     tcp_keepalive_probes - the number of probes that are sent and unacknowledged before the client considers the connection broken and notifies the application layer
 +  *     tcp_keepalive_time - the interval between the last data packet sent and the first keepalive probe
 +  *     tcp_keepalive_intvl - the interval between subsequent keepalive probes
 +  *     tcp_retries2 - the maximum number of times a packet is retransmitted before giving up
 +
 +
 +~~~bash
 +echo "6" > /proc/sys/net/ipv4/tcp_keepalive_time
 +echo "1" > /proc/sys/net/ipv4/tcp_keepalive_intvl
 +echo "10" > /proc/sys/net/ipv4/tcp_keepalive_probes
 +echo "3" > /proc/sys/net/ipv4/tcp_retries2
 +~~~
 +
 +Source : https://www.ibm.com/docs/en/db2/9.7?topic=ctkp-configuring-operating-system-tcpip-keepalive-parameters-high-availability-clients
 +
 +
 +~~~bash
 +ss -o
 +~~~
 +
 +
 +##  Process / diag tools
 +
 +Voir : 
 +* https://access.redhat.com/solutions/30453
 +
 +## Outils
 +
 +### TCP retransmissions
 +
 +Voir : 
 +* http://arthurchiao.art/blog/tcp-retransmission-may-be-misleading/
 +* net.ipv4.tcp_early_retrans
 +
 +Outils :
 +  * tcpretrans.bt (bpftrace)
 +  * tcpretrans [perf-tools](https://github.com/brendangregg/perf-tools/blob/master/net/tcpretrans)
 +  * tcpretrans.py [bpfcc-tools - iovisor/bcc](https://github.com/iovisor/bcc/blob/master/tools/tcpretrans.py)
 +
 +
 +Connaitre le rto_min et le rto_max
 +~~~
 +# grep ^Tcp /proc/net/snmp | column -t | cut -c1-99
 +Tcp:  RtoAlgorithm  RtoMin  RtoMax  MaxConn  ActiveOpens  PassiveOpens  AttemptFails  EstabResets
 +Tcp:  1             200     120000  -1       6834         964           161           4614
 +~~~
 +
 +~~~bash
 +yum install bpftrace
 +/usr/share/bcc/tools/tcpretrans
 +~~~
 +
 +~~~bash
 +timeout 60 ./tcpretrans | nl
 +~~~
 +
 +~~~bash
 +sar -n ETCP
 +sar -n TCP
 +~~~
 +
 +~~~
 +# netstat -s | egrep 'segments retransmited | segments send out'
 +    107428604792 segments send out
 +    47511527 segments retransmited
 +
 +# echo "$(( 47511527 * 10000 / 107428604792 ))"
 +4
 +~~~
 +
 +
 +https://www.ibm.com/support/pages/tracking-tcp-retransmissions-linux
 +
 +`tcpretransmits.sh`
 +~~~bash
 +#! /usr/bin/bash
 +
 +test -x /usr/sbin/tcpretrans.bt && TCPRETRANS=/usr/sbin/tcpretrans.bt
 +test -x /usr/share/bpftrace/tools/tcpretrans.bt && TCPRETRANS=/usr/share/bpftrace/tools/tcpretrans.bt
 +# https://github.com/brendangregg/perf-tools/blob/master/net/tcpretrans
 +test -x ./tcpretrans.pl && TCPRETRANS=./tcpretrans.pl
 +
 +OUT=/tmp/tcpretransmits.log
 +
 +if [ -z "$TCPRETRANS" ]; then
 +  echo "It looks like 'bpftrace' is not installed"
 +else
 +  date > $OUT
 +  netstat -s | awk '/segments sen. out$/ { R=$1; } /segments retransmit+ed$/ { printf("%.4f\n", ($1/R)*100); }' >> $OUT
 +  $TCPRETRANS | tee -a $OUT
 +  netstat -s | awk '/segments sen. out$/ { R=$1; } /segments retransmit+ed$/ { printf("%.4f\n", ($1/R)*100); }' >> $OUT
 +fi
 +~~~
 +
 +
 +Resolving The Problem \
 +TCP retransmissions are almost exclusively caused by failing network hardware, not applications or middleware.  Report the failing IP pairs to a network administrator.
 +
 +
 +
 +
 +
 +## Autres 
 +
 +** horodatages TCP **
 +https://access.redhat.com/documentation/fr-fr/red_hat_enterprise_linux/9/html/monitoring_and_managing_system_status_and_performance/benefits-of-tcp-timestamps_tuning-the-network-performance
 +
 +tcp_low_latency (Boolean; default: disabled; since Linux 2.4.21/2.6; obsolete since Linux 4.14)
 +
 +
 +~~~ini
 +net.ipv4.tcp_timestamps = 1
 +net.ipv4.tcp_window_scaling = 1
 +net.ipv4.tcp_sack = 1
 +net.ipv4.tcp_moderate_rcvbuf = 1
 +~~~
 +
 +----------------
 +
 +
 +~~~
 +# ip route get 192.168.100.11
 +192.168.100.11 dev virbr1 src 192.168.100.1 uid 1000 
 +    cache 
 +    
 +# ip route show dev virbr1
 +192.168.100.0/24 proto kernel scope link src 192.168.100.1
 +
 +
 +# ip route change dev virbr1 192.168.100.0/24 proto kernel scope link src 192.168.100.1 rto_min 8ms
 +~~~
 +
 +
 +
 +
 +
 +
 +
 +
  
Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki