User Tools
====== TCP Tuning ====== By default, Centos used the following TCP parameters: One can use "sysctl -w" to change these parameters. In order for changes to persist after reboot, one needs to edit "/etc/sysctl.conf" (typically as root). Click [[http://www.frozentux.net/ipsysctl-tutorial/ipsysctl-tutorial.html#AEN454|here]] for a description of all of these parameters. Other helpful sites: [[http://www.speedguide.net/articles/linux-tweaking-121 | SpeedGuide.net]] [[http://www.psc.edu/networking/projects/tcptune/ | PSC High Performance]] <code> $ sysctl -a net.ipv4.tcp_slow_start_after_idle = 1 net.ipv4.tcp_dma_copybreak = 4096 net.ipv4.tcp_workaround_signed_windows = 0 net.ipv4.tcp_base_mss = 512 net.ipv4.tcp_mtu_probing = 0 net.ipv4.tcp_abc = 0 net.ipv4.tcp_congestion_control = bic net.ipv4.tcp_tso_win_divisor = 3 net.ipv4.tcp_moderate_rcvbuf = 1 net.ipv4.tcp_no_metrics_save = 0 net.ipv4.ipfrag_max_dist = 64 net.ipv4.ipfrag_secret_interval = 600 net.ipv4.tcp_low_latency = 0 net.ipv4.tcp_frto = 0 net.ipv4.tcp_tw_reuse = 0 net.ipv4.icmp_ratemask = 6168 net.ipv4.icmp_ratelimit = 1000 net.ipv4.tcp_adv_win_scale = 2 net.ipv4.tcp_app_win = 31 net.ipv4.tcp_rmem = 4096 87380 4194304 net.ipv4.tcp_wmem = 4096 16384 4194304 net.ipv4.tcp_mem = 196608 262144 393216 net.ipv4.tcp_dsack = 1 net.ipv4.tcp_ecn = 0 net.ipv4.tcp_reordering = 3 net.ipv4.tcp_fack = 1 net.ipv4.tcp_orphan_retries = 0 net.ipv4.tcp_max_syn_backlog = 1024 net.ipv4.tcp_rfc1337 = 0 net.ipv4.tcp_stdurg = 0 net.ipv4.tcp_abort_on_overflow = 0 net.ipv4.tcp_tw_recycle = 0 net.ipv4.tcp_syncookies = 1 net.ipv4.tcp_fin_timeout = 60 net.ipv4.tcp_retries2 = 15 net.ipv4.tcp_retries1 = 3 net.ipv4.tcp_keepalive_intvl = 75 net.ipv4.tcp_keepalive_probes = 9 net.ipv4.tcp_keepalive_time = 7200 net.ipv4.tcp_max_tw_buckets = 180000 net.ipv4.tcp_max_orphans = 65536 net.ipv4.tcp_synack_retries = 5 net.ipv4.tcp_syn_retries = 5 net.ipv4.tcp_retrans_collapse = 1 net.ipv4.tcp_sack = 1 net.ipv4.tcp_window_scaling = 1 net.ipv4.tcp_timestamps = 1 net.core.netdev_budget = 300 net.core.somaxconn = 128 net.core.xfrm_larval_drop = 0 net.core.xfrm_acq_expires = 30 net.core.xfrm_aevent_rseqth = 2 net.core.xfrm_aevent_etime = 10 net.core.optmem_max = 20480 net.core.message_burst = 10 net.core.message_cost = 5 net.core.netdev_max_backlog = 1000 net.core.dev_weight = 64 net.core.rmem_default = 129024 net.core.wmem_default = 129024 net.core.rmem_max = 131071 net.core.wmem_max = 131071 </code> **net.core.rmem_max**: Maximum TCP receive window. FasterData recommends 16MB (**16777216**) for 10G paths with a few parallel streams, or 32MB for very long end-to-end 10G or 40G paths. **net.core.wmem_max**: Maximum TCP Send Window. FasterData recommends 16MB (**16777216**) for 10G paths with a few parallel streams, or 32MB for very long end-to-end 10G or 40G paths. **net.ipv4.tcp_rmem**: Memory reserved for TCP receive buffers (reserved memory per connection default). Format: min, default, and max number of bytes to use. FasterData recommends **4096 87380 16777216** and only changing the 3rd parameter to use more than 16MB. **net.ipv4.tcp_wmem**: Memory reserved for TCP send buffers (reserved memory per connection default). Format: min, default, and max number of bytes to use. FasterData recommends **4096 87380 16777216** and only changing the 3rd parameter to use more than 16MB. **net.core.netdev_max_backlog**: FasterData recommends **30000**, and increasing for 10G or faster links. **net.ipv4.tcp_congestion_control**: according to [[http://fasterdata.es.net/fasterdata/host-tuning/linux/expert/|FasterData]], **htcp** (Hamilton TCP) or **cubic** is better for long, high speed links. <code> sysctl -w net.core.rmem_max=16777216 sysctl -w net.core.wmem_max=16777216 sysctl -w net.ipv4.tcp_rmem=4096 87380 16777216 sysctl -w net.ipv4.tcp_wmem=4096 87380 16777216 sysctl -w net.core.netdev_max_backlog=30000 sysctl -w net.ipv4.tcp_congestion_control=htcp </code> [[http://www.speedguide.net/articles/linux-tweaking-121|SpeedGuide]] also recommends the following settings: <code> sysctl -w net.ipv4.tcp_sack = 1 sysctl -w net.ipv4.tcp_window_scaling = 1 sysctl -w net.ipv4.tcp_timestamps = 0 </code> Although setting the tcp_timestamps = 0 is controversial (and we chose to leave it enabled): FasterData points out that some alternative congestion control protocols require accurate timestamps. The Ethernet cards we are testing with are: <code> eth1: Tigon3 [partno(BCM95704A7) rev 2003 PHY(5704)] (PCI:66MHz:64-bit) 10/100/1000Base-T Ethernet 00:e0:81:2c:5e:bd eth1: RXcsums[1] LinkChgREG[0] MIirq[0] ASF[0] WireSpeed[1] TSOcap[1] eth1: dma_rwctrl[763f0000] dma_mask[64-bit] </code> Also, we will experiment with the MTU (using Jumbo frames of size 9000): <code> ifconfig eth1 mtu 9000 </code> And to test: <code> ping -M do -s 8992 192.168.0.3 </code>
