{"id":25656,"date":"2026-11-19T14:02:33","date_gmt":"2026-11-19T14:02:33","guid":{"rendered":"https:\/\/atalnetworks.com\/?p=25656"},"modified":"2026-11-19T14:02:33","modified_gmt":"2026-11-19T14:02:33","slug":"linux-network-tuning-for-10gbps-dedicated-servers","status":"publish","type":"post","link":"https:\/\/atalnetworks.com\/fr\/linux-network-tuning-for-10gbps-dedicated-servers\/","title":{"rendered":"R\u00e9glage du r\u00e9seau Linux pour les serveurs d\u00e9di\u00e9s de 10\u00a0Gbit\/s\u00a0: mesurez avant de modifier"},"content":{"rendered":"<p><span style=\"font-weight: 400;\">Linux network tuning for 10Gbps starts with throughput, latency, packet-drop, and CPU measurements. A useful change addresses a measured limit, improves the real workload, and can be reversed. Larger buffers, more queues, or a different TCP algorithm can help some transfers and make other traffic worse.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">We use a repeatable process in this guide: record the starting state, locate the likely limit, test one change, and compare the result. The goal is a server that meets its workload targets without extra delay, wasted memory, or fragile configuration.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">What should you measure before tuning a 10Gbps Linux server?<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Measure useful throughput and the resources consumed to deliver it. Port speed alone does not show whether the limit sits in the host, application, remote endpoint, or network path.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">If you have not located the likely layer, start with our guide to <\/span><a href=\"https:\/\/atalnetworks.com\/why-is-my-10gbps-server-slow\/\"><span style=\"font-weight: 400;\">diagnosing a slow 10Gbps server<\/span><\/a><span style=\"font-weight: 400;\">. Continue here when the evidence points to Linux host processing or TCP behavior.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Set a goal for the real workload<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Write down what needs to improve. For a backup server, that could be completion time. For an API server, request latency during busy periods may matter more than peak transfer speed.<\/span><\/p>\n<p>&nbsp;<\/p>\n<table>\n<thead>\n<tr>\n<th><b>Measurement<\/b><\/th>\n<th><b>What to record<\/b><\/th>\n<th><b>Why it matters<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">Receiver throughput<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Delivered Gbit\/s or application MB\/s<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Measures useful delivery<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Idle and loaded latency<\/span><\/td>\n<td><span style=\"font-weight: 400;\">RTT and application response times<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Shows delay added during traffic<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Retransmissions<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Change during each TCP test<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Helps identify delivery problems<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Interface counters<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Errors and drops before and after<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Helps locate device or link trouble<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">CPU per core<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Busy cores and softirq activity<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Exposes limits hidden by average CPU use<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Memory<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Socket use and host memory pressure<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Checks the cost of buffer changes<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Application result<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Completion time or response-time distribution<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Shows whether the change helped users<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Set acceptable limits before testing. We recommend recording the median and spread across at least three comparable trials. This is a practical comparison method, not a claim that three runs describe every traffic condition.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Record the host and path<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">The examples use Bash and common Linux tools: iproute2, ethtool, procps, sysstat, and iperf3. Install missing tools through your distribution&#8217;s package manager. Flags, counters, and supported settings depend on the installed versions and driver.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Run these read-only checks on each endpoint:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">cat \/etc\/os-release<\/span><\/p>\n<p><span style=\"font-weight: 400;\">uname -r<\/span><\/p>\n<p><span style=\"font-weight: 400;\">iperf3 &#8211;version<\/span><\/p>\n<p><span style=\"font-weight: 400;\">ethtool &#8211;version<\/span><\/p>\n<p><span style=\"font-weight: 400;\">ip -br address<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">In the commands below, replace the example values before use. <\/span><span style=\"font-weight: 400;\">192.0.2.20<\/span><span style=\"font-weight: 400;\"> is a documentation address, not a public test server.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">peer=<\/span><span style=\"font-weight: 400;\">192<\/span><span style=\"font-weight: 400;\">.<\/span><span style=\"font-weight: 400;\">0<\/span><span style=\"font-weight: 400;\">.<\/span><span style=\"font-weight: 400;\">2<\/span><span style=\"font-weight: 400;\">.<\/span><span style=\"font-weight: 400;\">20<\/span><\/p>\n<p><span style=\"font-weight: 400;\">iface=enp1s0<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">ip route get <\/span><span style=\"font-weight: 400;\">&#8220;$peer&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">ethtool <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">ethtool -i <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Check that the route uses the expected interface and source address. A bond, bridge, VLAN, or tunnel may require inspection of both the logical interface and the underlying physical port.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Record NIC model, firmware, negotiated link speed, and the peer&#8217;s capacity. Check the NIC&#8217;s PCIe connection against its hardware requirements if the host remains suspect. Our <\/span><a href=\"https:\/\/atalnetworks.com\/how-to-choose-a-10gbps-dedicated-server\/\"><span style=\"font-weight: 400;\">10Gbps server selection guide<\/span><\/a><span style=\"font-weight: 400;\"> covers the wider hardware decision.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Save state before making changes<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Create a private working directory in the same Bash terminal:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">umask <\/span><span style=\"font-weight: 400;\">077<\/span><\/p>\n<p><span style=\"font-weight: 400;\">run_dir=$(mktemp -d <\/span><span style=\"font-weight: 400;\">&#8220;$PWD\/network-tuning.XXXXXX&#8221;<\/span><span style=\"font-weight: 400;\">)<\/span><\/p>\n<p><span style=\"font-weight: 400;\">printf <\/span><span style=\"font-weight: 400;\">&#8216;Test records: %s\\n&#8217; &#8220;$run_dir&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">date -Is &gt; <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/start-time.txt&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Keep its absolute path in your change record. Shell variables disappear when the session ends.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Use console or out-of-band access before changing an interface that carries SSH. Confirm who manages the settings: a network profile, TuneD, a boot service, or a deployment tool may reapply them. Schedule tests so that they do not disrupt customer traffic.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Run single-stream, parallel-stream, and reverse tests<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Test one TCP stream, several streams, and both directions before changing Linux network settings. Each result answers a different question.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Use two separate, capable hosts. Start the listener on the intended receiving endpoint and restrict TCP port 5201 to the test peer through your normal firewall controls:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">iperf3 -s<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">On the client, run these tests one at a time:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">iperf3 -c <\/span><span style=\"font-weight: 400;\">&#8220;$peer&#8221; <\/span><span style=\"font-weight: 400;\">-P <\/span><span style=\"font-weight: 400;\">1 <\/span><span style=\"font-weight: 400;\">-t <\/span><span style=\"font-weight: 400;\">30 <\/span><span style=\"font-weight: 400;\">-J &gt; <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/forward-one.json&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">iperf3 -c <\/span><span style=\"font-weight: 400;\">&#8220;$peer&#8221; <\/span><span style=\"font-weight: 400;\">-P <\/span><span style=\"font-weight: 400;\">4 <\/span><span style=\"font-weight: 400;\">-t <\/span><span style=\"font-weight: 400;\">30 <\/span><span style=\"font-weight: 400;\">-J &gt; <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/forward-four.json&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">iperf3 -c <\/span><span style=\"font-weight: 400;\">&#8220;$peer&#8221; <\/span><span style=\"font-weight: 400;\">-P <\/span><span style=\"font-weight: 400;\">1 <\/span><span style=\"font-weight: 400;\">-t <\/span><span style=\"font-weight: 400;\">30 <\/span><span style=\"font-weight: 400;\">-R -J &gt; <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/reverse-one.json&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">iperf3 -c <\/span><span style=\"font-weight: 400;\">&#8220;$peer&#8221; <\/span><span style=\"font-weight: 400;\">-P <\/span><span style=\"font-weight: 400;\">4 <\/span><span style=\"font-weight: 400;\">-t <\/span><span style=\"font-weight: 400;\">30 <\/span><span style=\"font-weight: 400;\">-R -J &gt; <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/reverse-four.json&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">-P<\/span><span style=\"font-weight: 400;\"> selects the stream count. <\/span><span style=\"font-weight: 400;\">-R<\/span><span style=\"font-weight: 400;\"> makes the server send to the client. <\/span><span style=\"font-weight: 400;\">-J<\/span><span style=\"font-weight: 400;\"> saves JSON output. Use a new filename for each repeat.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Beginning with iperf3 3.16, each test stream runs in its own thread. Record the version on both endpoints because tool behavior can affect CPU use and results.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">If four streams outperform one, investigate the single-flow limit. Do not treat the larger total as proof that one download can achieve the same rate.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Keep the peer, duration, direction, stream count, and background load as consistent as possible. A 30-second interval at a constant 10Gbps represents 37.5 GB at the raw-rate calculation, before protocol overhead. Actual transferred and billable volume depends on the test and service terms.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Measure while the connection is active<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Use another terminal on each host. Set its <\/span><span style=\"font-weight: 400;\">peer<\/span><span style=\"font-weight: 400;\"> and <\/span><span style=\"font-weight: 400;\">iface<\/span><span style=\"font-weight: 400;\"> variables for that host:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">mpstat -P ALL <\/span><span style=\"font-weight: 400;\">1 10<\/span><\/p>\n<p><span style=\"font-weight: 400;\">ss -tinm dst <\/span><span style=\"font-weight: 400;\">&#8220;$peer&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">ss<\/span><span style=\"font-weight: 400;\"> can display TCP information and socket memory. Inspect fields such as RTT, retransmissions, congestion window, and send or receive limits when your version exposes them. A field is evidence to interpret, not a diagnosis by itself.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Compare ping before and during the load as a rough path-delay check. Also measure application latency. ICMP replies do not replace HTTP, database, or game-server response measurements.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">For the complete test setup and output interpretation, use our <\/span><a href=\"https:\/\/atalnetworks.com\/test-10gbps-server-iperf3\/\"><span style=\"font-weight: 400;\">10Gbps iperf3 testing guide<\/span><\/a><span style=\"font-weight: 400;\">.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Match the symptom to the next check<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Choose a tuning test only after the measurements point to a likely limit. The same throughput symptom can have several causes.<\/span><\/p>\n<p>&nbsp;<\/p>\n<table>\n<thead>\n<tr>\n<th><b>Measured symptom<\/b><\/th>\n<th><b>Check next<\/b><\/th>\n<th><b>Possible action<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">One stream is slow; several streams are faster<\/span><\/td>\n<td><span style=\"font-weight: 400;\">RTT, socket state, one busy core, peer limits<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Test the specific flow limit<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">One receive-processing core stays busy<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Queue activity, IRQ distribution, CPU topology<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Assess queue or CPU placement<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">NIC receive-drop counters rise<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Driver counter definitions, ring depth, CPU drain rate<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Test a supported ring change<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Kernel backlog-drop counters rise<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Per-core processing, backlog, processing budgets<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Test the matching software limit<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Latency rises sharply under load<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Local and path queues, active qdisc<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Investigate queueing and pacing<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Physical error counters rise<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Cable, optic, port, driver, switch evidence<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Investigate the link<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">iperf3 is healthy; file transfers remain slow<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Storage, encryption, application workers, remote storage<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Test the application path<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Only one destination is slow<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Remote host and network path<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Compare another capable endpoint<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Collect absolute counters before each test:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">ip -s link show dev <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221; <\/span><span style=\"font-weight: 400;\">&gt; <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/link-before.txt&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">ethtool -S <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221; <\/span><span style=\"font-weight: 400;\">&gt; <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/nic-before.txt&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">nstat -asz &gt; <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/nstat-before.txt&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">cat \/proc\/net\/softnet_stat &gt; <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/softnet-before.txt&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Repeat after the test with <\/span><span style=\"font-weight: 400;\">after<\/span><span style=\"font-weight: 400;\"> filenames. <\/span><span style=\"font-weight: 400;\">nstat -asz<\/span><span style=\"font-weight: 400;\"> prints absolute counters, includes zero values, and avoids updating its history file.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Calculate changes over the same interval. A nonzero lifetime counter may describe an old event. Host-wide counters may also include unrelated traffic, so compare them with the workload and device evidence.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Use RTT and BDP to assess TCP buffer needs<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Use bandwidth-delay product, or BDP, to estimate the data that must be in flight at a target rate. Then inspect socket behavior before testing larger buffers.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">The calculation is:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><b>BDP in bytes = target rate in bits per second \u00d7 RTT in seconds \u00f7 8<\/b><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">At a target rate of 10,000,000,000 bits per second:<\/span><\/p>\n<p>&nbsp;<\/p>\n<table>\n<thead>\n<tr>\n<th><b>RTT<\/b><\/th>\n<th><b>Calculated data in flight<\/b><\/th>\n<th><b>Decimal MB<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">1 ms<\/span><\/td>\n<td><span style=\"font-weight: 400;\">1,250,000 bytes<\/span><\/td>\n<td><span style=\"font-weight: 400;\">1.25 MB<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">10 ms<\/span><\/td>\n<td><span style=\"font-weight: 400;\">12,500,000 bytes<\/span><\/td>\n<td><span style=\"font-weight: 400;\">12.5 MB<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">50 ms<\/span><\/td>\n<td><span style=\"font-weight: 400;\">62,500,000 bytes<\/span><\/td>\n<td><span style=\"font-weight: 400;\">62.5 MB<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">100 ms<\/span><\/td>\n<td><span style=\"font-weight: 400;\">125,000,000 bytes<\/span><\/td>\n<td><span style=\"font-weight: 400;\">125 MB<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">These are calculated values, not benchmark results or direct sysctl recommendations. TCP window behavior, memory accounting, loss, flow sharing, and endpoint capacity still matter. Red Hat uses BDP as part of its TCP buffer-sizing procedure and warns about memory and latency costs.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Understand the settings you are testing<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Read the current values:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">sysctl net.ipv4.tcp_rmem net.ipv4.tcp_wmem<\/span><\/p>\n<p><span style=\"font-weight: 400;\">sysctl net.ipv4.tcp_moderate_rcvbuf<\/span><\/p>\n<p><span style=\"font-weight: 400;\">sysctl net.core.rmem_max net.core.wmem_max<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">tcp_rmem<\/span><span style=\"font-weight: 400;\"> and <\/span><span style=\"font-weight: 400;\">tcp_wmem<\/span><span style=\"font-weight: 400;\"> contain minimum, initial, and maximum values. The maximum controls the automatically selected TCP buffer size. Receive autotuning uses <\/span><span style=\"font-weight: 400;\">tcp_moderate_rcvbuf<\/span><span style=\"font-weight: 400;\">.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">An application that sets <\/span><span style=\"font-weight: 400;\">SO_RCVBUF<\/span><span style=\"font-weight: 400;\"> or <\/span><span style=\"font-weight: 400;\">SO_SNDBUF<\/span><span style=\"font-weight: 400;\"> changes the sizing behavior for that socket. Inspect its configuration and applicable core limits before assuming the TCP autotuning ceiling controls the application. Linux also includes accounting overhead in socket-buffer handling.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Do not raise minimums and initial sizes merely because the NIC supports 10Gbps. A higher maximum is a ceiling, not that amount of memory allocated to every socket. Connection count and memory pressure still need review.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Example: test a receive-buffer ceiling<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Use this example only when evidence points to a receive-buffer limit and the host has room for the trial. Suppose you have selected a <\/span><b>64 MiB ceiling<\/b><span style=\"font-weight: 400;\"> for a controlled test. That is an example value, not a 10Gbps preset.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">On the receiving host, save the original triplet:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">sysctl net.ipv4.tcp_rmem &gt; <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/tcp-rmem.before.conf&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">read <\/span><span style=\"font-weight: 400;\">-r rx_min rx_initial rx_max &lt; &lt;(sysctl -n net.ipv4.tcp_rmem)<\/span><\/p>\n<p><span style=\"font-weight: 400;\">trial_rx_max=<\/span><span style=\"font-weight: 400;\">67108864<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">If the read succeeded and the current maximum is smaller, test only the maximum:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">if <\/span><span style=\"font-weight: 400;\">[[ <\/span><span style=\"font-weight: 400;\">&#8220;$rx_min&#8221; <\/span><span style=\"font-weight: 400;\">=~ ^[<\/span><span style=\"font-weight: 400;\">0-9<\/span><span style=\"font-weight: 400;\">]+$ &amp;&amp;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span><span style=\"font-weight: 400;\">&#8220;$rx_initial&#8221; <\/span><span style=\"font-weight: 400;\">=~ ^[<\/span><span style=\"font-weight: 400;\">0-9<\/span><span style=\"font-weight: 400;\">]+$ &amp;&amp;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span><span style=\"font-weight: 400;\">&#8220;$rx_max&#8221; <\/span><span style=\"font-weight: 400;\">=~ ^[<\/span><span style=\"font-weight: 400;\">0-9<\/span><span style=\"font-weight: 400;\">]+$ ]] &amp;&amp;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">\u00a0\u00a0\u00a0(( rx_max &lt; trial_rx_max )); <\/span><span style=\"font-weight: 400;\">then<\/span><\/p>\n<p><span style=\"font-weight: 400;\">\u00a0\u00a0\u00a0\u00a0<\/span><span style=\"font-weight: 400;\">sudo sysctl -w <\/span><span style=\"font-weight: 400;\">&#8220;net.ipv4.tcp_rmem=$rx_min $rx_initial $trial_rx_max&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">else<\/span><\/p>\n<p><span style=\"font-weight: 400;\">\u00a0\u00a0\u00a0\u00a0<\/span><span style=\"font-weight: 400;\">printf <\/span><span style=\"font-weight: 400;\">&#8216;%s\\n&#8217; &#8216;No change: inspect the current values.&#8217;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">fi<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">sysctl net.ipv4.tcp_rmem<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Repeat the same transfer and inspect the socket, throughput, latency, and memory result. Restore the original value after the trial unless the result justifies keeping it. First check that the live setting still belongs to your trial and no other administrator changed it:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">cat <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/tcp-rmem.before.conf&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">sysctl net.ipv4.tcp_rmem<\/span><\/p>\n<p><span style=\"font-weight: 400;\">sudo sysctl -p <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/tcp-rmem.before.conf&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">sysctl net.ipv4.tcp_rmem<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">sysctl -p<\/span><span style=\"font-weight: 400;\"> loads the named file. Use the saved file for this host and trial only. Keep the send-buffer test separate, on the sender, with its own saved <\/span><span style=\"font-weight: 400;\">tcp_wmem<\/span><span style=\"font-weight: 400;\"> value.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Tune NIC rings and kernel backlogs only for matching drops<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Increase queue capacity only when the matching counters and workload support that test. A larger queue cannot make a persistently overloaded CPU process packets faster.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Keep queue types separate<\/span><\/h3>\n<table>\n<thead>\n<tr>\n<th><b>Queue or buffer<\/b><\/th>\n<th><b>What it handles<\/b><\/th>\n<th><b>Inspection<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">NIC RX\/TX rings<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Packet descriptors used by the device and driver<\/span><\/td>\n<td><span style=\"font-weight: 400;\">ethtool -g<\/span><span style=\"font-weight: 400;\">, driver statistics<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Kernel input backlog<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Packets awaiting software processing where this path applies<\/span><\/td>\n<td><span style=\"font-weight: 400;\">softnet_stat<\/span><span style=\"font-weight: 400;\">, relevant sysctl values<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">TCP socket buffers<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Data associated with a connection<\/span><\/td>\n<td><span style=\"font-weight: 400;\">ss -tinm<\/span><span style=\"font-weight: 400;\">, TCP settings<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Listen queues<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Pending connection work<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Listener state and application metrics<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Egress qdisc<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Packet scheduling on transmit<\/span><\/td>\n<td><span style=\"font-weight: 400;\">tc -s qdisc show<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Do not use a larger listen backlog as a general remedy for a slow established transfer. First identify whether the problem concerns connection acceptance or data delivery.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Example: test RX ring depth<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Inspect and save the current state:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">ethtool -g <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221; <\/span><span style=\"font-weight: 400;\">| tee <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/rings-before.txt&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">ethtool -S <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221; <\/span><span style=\"font-weight: 400;\">&gt; <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/ring-counters-before.txt&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Suppose the current RX ring is <\/span><b>512 descriptors<\/b><span style=\"font-weight: 400;\">, the driver supports <\/span><b>1024<\/b><span style=\"font-weight: 400;\">, and its documented counters show receive-ring pressure during the test. In a maintenance window, a supported trial is:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">sudo ethtool -G <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221; <\/span><span style=\"font-weight: 400;\">rx <\/span><span style=\"font-weight: 400;\">1024<\/span><\/p>\n<p><span style=\"font-weight: 400;\">ethtool -g <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Repeat the workload and compare drops, throughput, latency, and CPU use. If the trial does not meet the goal, and the current state still matches your trial, restore this example&#8217;s original value:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">sudo ethtool -G <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221; <\/span><span style=\"font-weight: 400;\">rx <\/span><span style=\"font-weight: 400;\">512<\/span><\/p>\n<p><span style=\"font-weight: 400;\">ethtool -g <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Use the values from your own device. Some drivers can briefly interrupt traffic during ring changes. Recheck related device state after the operation.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Check software backlog and processing budgets<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Read the relevant controls:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">sysctl net.core.netdev_max_backlog<\/span><\/p>\n<p><span style=\"font-weight: 400;\">sysctl net.core.netdev_budget net.core.netdev_budget_usecs<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">netdev_max_backlog<\/span><span style=\"font-weight: 400;\"> limits the input backlog. <\/span><span style=\"font-weight: 400;\">netdev_budget<\/span><span style=\"font-weight: 400;\"> and <\/span><span style=\"font-weight: 400;\">netdev_budget_usecs<\/span><span style=\"font-weight: 400;\"> limit work within a packet-processing polling cycle.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Compare per-CPU <\/span><span style=\"font-weight: 400;\">softnet_stat<\/span><span style=\"font-weight: 400;\"> snapshots. On commonly documented layouts, the second field records backlog drops and the third records processing-budget pressure. Confirm the layout for the running kernel before parsing its hexadecimal fields.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">If a matching counter rises, test one supported change at a time and save its old value first. More processing time can compete with application work. More backlog can add waiting time. Restore the setting if the trade-off fails your service targets.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Check RSS, IRQ distribution, and NUMA placement<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Inspect CPU placement when traffic overloads one processing core or a small group of queues. Low total CPU use does not rule out a per-core limit.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Receive Side Scaling, or RSS, maps flows to hardware receive queues. A flow normally stays on one queue; adding queues does not split that flow across every CPU. RPS steers receive processing in software, RFS considers application placement, and XPS helps select transmit queues.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Read the queue and interrupt state:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">ethtool -l <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">ethtool -x <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">cat \/proc\/interrupts<\/span><\/p>\n<p><span style=\"font-weight: 400;\">lscpu -e<\/span><\/p>\n<p><span style=\"font-weight: 400;\">systemctl status irqbalance<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">For a physical PCIe NIC, also inspect:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">cat <\/span><span style=\"font-weight: 400;\">&#8220;\/sys\/class\/net\/$iface\/device\/numa_node&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">A value of <\/span><span style=\"font-weight: 400;\">-1<\/span><span style=\"font-weight: 400;\"> means the node is unknown. Logical interfaces may not expose that path.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Use the following sequence:<\/span><\/p>\n<p>&nbsp;<\/p>\n<ol>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Compare per-core load with queue and IRQ activity during the test.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Check whether irqbalance already manages the assignment.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Compare NIC locality with the CPUs and application workers involved.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Test one placement or supported channel change only if the evidence warrants it.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Re-read queue and IRQ identities after the change.<\/span><\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Do not paste a fixed CPU mask from another machine. Save the original assignments and manager state before any manual trial. If you restore channel counts, check the resulting IRQ identifiers before restoring placement. irqbalance may overwrite manual placement.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Test congestion control and qdiscs as separate changes<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Compare TCP algorithms only under matched conditions, and inspect packet scheduling as a separate part of the test. Changing both at once makes the result harder to explain.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Read the current state:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">sysctl net.ipv4.tcp_available_congestion_control<\/span><\/p>\n<p><span style=\"font-weight: 400;\">sysctl net.ipv4.tcp_congestion_control<\/span><\/p>\n<p><span style=\"font-weight: 400;\">tc -s qdisc show dev <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">The available list shows registered algorithms. The default does not establish which algorithm every existing socket uses.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">For a forward iperf3 test, <\/span><span style=\"font-weight: 400;\">-C<\/span><span style=\"font-weight: 400;\"> can request a sender-side algorithm without changing the host default. If both choices are available and permitted, compare:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">iperf3 -c <\/span><span style=\"font-weight: 400;\">&#8220;$peer&#8221; <\/span><span style=\"font-weight: 400;\">-P <\/span><span style=\"font-weight: 400;\">1 <\/span><span style=\"font-weight: 400;\">-t <\/span><span style=\"font-weight: 400;\">30 <\/span><span style=\"font-weight: 400;\">-C cubic -J &gt; <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/cubic-one.json&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">iperf3 -c <\/span><span style=\"font-weight: 400;\">&#8220;$peer&#8221; <\/span><span style=\"font-weight: 400;\">-P <\/span><span style=\"font-weight: 400;\">1 <\/span><span style=\"font-weight: 400;\">-t <\/span><span style=\"font-weight: 400;\">30 <\/span><span style=\"font-weight: 400;\">-C bbr -J &gt; <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/bbr-one.json&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Confirm the selected algorithm and inspect errors. Use repeated trials and record the installed kernel&#8217;s implementation. \u201cBBR\u201d alone does not describe every version or downstream build.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Keep the qdisc unchanged during this comparison. Assess receiver throughput, loaded latency, retransmissions, and other active traffic. Neither algorithm wins every workload.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">A qdisc schedules egress packets. The <\/span><span style=\"font-weight: 400;\">fq<\/span><span style=\"font-weight: 400;\"> scheduler supports per-flow pacing, but changing a default qdisc setting is not evidence that the active interface hierarchy changed.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Before any qdisc trial, prepare commands that can recreate the original hierarchy, including classes, filters, and options. A text listing is a record, not an automatic restore script. Do not replace production traffic policy with a generic <\/span><span style=\"font-weight: 400;\">tc qdisc replace<\/span><span style=\"font-weight: 400;\"> command.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Check offloads and interrupt coalescing against the workload<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Keep working offloads unless a specific workload, driver issue, or measurement gives a reason to test a change. Disabling every feature can add CPU work.<\/span><\/p>\n<p>&nbsp;<\/p>\n<table>\n<thead>\n<tr>\n<th><b>Feature<\/b><\/th>\n<th><b>Main function<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">TSO<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Lets supported hardware segment TCP data<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">GSO<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Supports segmentation later in the transmit path<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">GRO<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Combines compatible received packets for processing<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Checksum offload<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Moves supported checksum work to the device<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Feature dependencies matter. For example, TCP segmentation offload depends on partial checksum support.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Read and save the full state:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">ethtool -k <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221; <\/span><span style=\"font-weight: 400;\">| tee <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/offloads-before.txt&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">ethtool -c <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221; <\/span><span style=\"font-weight: 400;\">| tee <\/span><span style=\"font-weight: 400;\">&#8220;$run_dir\/coalesce-before.txt&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">If you have a specific reason to test GRO, confirm that it is currently on and can be changed. A temporary off trial is:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">sudo ethtool -K <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221; <\/span><span style=\"font-weight: 400;\">gro off<\/span><\/p>\n<p><span style=\"font-weight: 400;\">ethtool -k <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Repeat the workload. Compare CPU cost and application latency as well as throughput. If this trial started with GRO on, restore it after checking the live state:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">sudo ethtool -K <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221; <\/span><span style=\"font-weight: 400;\">gro on<\/span><\/p>\n<p><span style=\"font-weight: 400;\">ethtool -k <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Use the full feature listing to check for dependent changes. Entries marked <\/span><span style=\"font-weight: 400;\">[fixed]<\/span><span style=\"font-weight: 400;\"> cannot be changed through that control.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Interrupt coalescing reduces interrupt frequency by batching notifications. This can reduce processing cost but may add delay. Inspect adaptive settings before testing timing or frame thresholds. Save all changed values and repeat the latency test after restoration.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Local packet captures can also reflect offload processing rather than packets exactly as they appear on the wire. Check capture location and driver behavior before treating large captured packets or checksum reports as link faults.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Check path MTU before testing jumbo frames<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Use a larger MTU only on a path that supports it. A 10Gbps port does not require an MTU of 9000.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">The interface MTU limits packet size on that interface. Path MTU concerns the usable size across the route. TCP MSS describes TCP data size within a segment. Tunnels add headers, which can reduce the size available to inner traffic.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Read the current interface value:<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">ip link show dev <\/span><span style=\"font-weight: 400;\">&#8220;$iface&#8221;<\/span><\/p>\n<p><span style=\"font-weight: 400;\">cat <\/span><span style=\"font-weight: 400;\">&#8220;\/sys\/class\/net\/$iface\/mtu&#8221;<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">For a controlled private network, confirm support on both endpoints, switches, logical interfaces, and any intermediate devices before a trial. Driver limits differ; the ixgbe documentation, for example, lists adapter-specific MTU and configuration details.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Use this order:<\/span><\/p>\n<p>&nbsp;<\/p>\n<ol>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Record the current MTU and persistent profile.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Confirm the complete relevant path supports the trial size.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Arrange console access and a maintenance window.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Change only the intended path and verify connectivity.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Repeat the workload and compare CPU use, throughput, and latency.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Restore the saved value if the trial fails.<\/span><\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Do not change an internet-facing interface to 9000 because two local devices support it. Also, <\/span><span style=\"font-weight: 400;\">tcp_mtu_probing<\/span><span style=\"font-weight: 400;\"> controls TCP packetization-layer probing; it does not turn an incompatible path into a jumbo-frame path.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Repeat the application test and keep only useful changes<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Keep a tuning change only when the real workload improves within its resource and latency limits. A better iperf3 total is one part of that decision.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Consider a backup job that reads data, encrypts it, sends it, and writes it at the destination. Network-only throughput can improve while the job remains limited by either storage system or encryption work. Test the complete job before making another kernel change.<\/span><\/p>\n<p>&nbsp;<\/p>\n<table>\n<thead>\n<tr>\n<th><b>Workload<\/b><\/th>\n<th><b>Result to check after tuning<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">Backup or replication<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Completion time, storage activity, other services&#8217; latency<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">HTTPS downloads<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Delivered rate, TLS CPU use, application workers, client capacity<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">API or database service<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Request rate, errors, and high-percentile response time<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">VPN gateway<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Encrypted throughput, CPU use, tunnel MTU, forwarding behavior<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Game or voice service<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Application delay, packet delivery, and CPU scheduling<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">For each change, complete a record:<\/span><\/p>\n<p>&nbsp;<\/p>\n<table>\n<thead>\n<tr>\n<th><b>Field<\/b><\/th>\n<th><b>Entry<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">Goal<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Workload and acceptance limits<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Evidence<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Counter changes and observed limit<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Saved state<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Host, interface, namespace, values, files, manager<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Trial<\/span><\/td>\n<td><span style=\"font-weight: 400;\">One setting and the reason for testing it<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Baseline<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Median and spread across comparable runs<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">After change<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Throughput, latency, drops, CPU, memory, application result<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Decision<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Keep, restore, or investigate another layer<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Restore check<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Original state restored and normal traffic checked<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Use a fresh record for a new hypothesis. A small gain within normal run-to-run variation is weak evidence for a permanent change.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Keep other traffic types in scope<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">TCP endpoint tests do not establish the performance of a forwarding gateway, a UDP service, or QUIC. Test packet size, packet rate, concurrency, encryption, and the actual application path for those workloads. TCP congestion-control settings do not select the congestion algorithm used by a QUIC implementation.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">For containers or network namespaces, identify where the setting applies. Record whether a command reads the host, a namespace, or a virtual interface before comparing values.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Restore failed changes and make proven settings persistent<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Restore the exact saved state for the setting you changed. Then check whether a persistent file or manager can reapply the trial.<\/span><\/p>\n<p>&nbsp;<\/p>\n<table>\n<thead>\n<tr>\n<th><b>Setting<\/b><\/th>\n<th><b>Save before testing<\/b><\/th>\n<th><b>Restore and check<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">sysctl<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Exact targeted value or triplet<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Load the saved value and re-read it<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Rings<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Current and supported ring state<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Restore supported original values and inspect related state<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Channels and RSS<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Counts, mapping, and related IRQ state<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Restore counts and mapping; recheck IRQ identities<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">IRQ placement<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Assignments and irqbalance state<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Restore the correct current IRQs and manager state<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Offloads<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Full supported feature state<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Restore changed features and inspect dependencies<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Coalescing<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Timing, thresholds, adaptive state<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Restore recorded settings and repeat latency checks<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">MTU<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Live value and network profile<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Restore both as needed and check connectivity<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">TCP algorithm<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Default and application overrides<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Restore the relevant setting and test new connections<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Qdisc<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Hierarchy plus reconstruction commands<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Recreate the original policy and inspect it<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">If live values differ from your trial, stop and check ownership before restoring anything. Another administrator or configuration service may have changed them.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Persist only the accepted settings<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">For an accepted sysctl change, place the reviewed assignment in a dedicated file under <\/span><span style=\"font-weight: 400;\">\/etc\/sysctl.d\/<\/span><span style=\"font-weight: 400;\">, using the site&#8217;s naming and configuration-management rules. Apply that file explicitly and re-read the value. Avoid reloading every sysctl file just to apply one test result.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">On systemd-based systems, sysctl.d files can set values at boot, and file ordering affects which assignment wins. Keep the reason, baseline, accepted result, owner, and rollback value beside the change record.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Use your distribution&#8217;s supported network-profile mechanism for persistent ethtool or MTU settings. Check what happens after an interface restart, driver reload, and planned reboot.<\/span><\/p>\n<p>&nbsp;<\/p>\n<p><span style=\"font-weight: 400;\">Deleting a trial file alone does not restore a live sysctl value. Restore the runtime value as well, and remove or correct the entry that would reapply it.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Frequently Asked Questions<\/span><\/h2>\n<h3><span style=\"font-weight: 400;\">Do modern Linux servers need manual tuning for 10Gbps?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Some workloads need a targeted change; others already meet their goals with current settings. Start with a repeatable test and live evidence. If throughput, latency, drops, CPU, and memory meet the service targets, keep the configuration and monitor it. Port speed alone is not a reason to replace Linux defaults.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Why is one TCP stream slower than several streams?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Multiple streams can use more processing resources and have separate TCP state. A single stream may face a per-core, window, congestion, application, or peer limit. Compare both results, then inspect the active socket and CPU distribution. The multi-stream total does not establish the speed available to one connection.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">How much TCP buffer does a 10Gbps connection need?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">There is no single value for every 10Gbps path. At 50 ms RTT, the calculated BDP at a 10Gbps target is 62.5 MB of data in flight. Actual buffer settings also depend on TCP behavior, accounting overhead, application options, memory pressure, and flow count. Use the calculation to guide investigation, then test.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Should we use BBR or keep CUBIC?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Keep the current algorithm until a matched comparison supports a change. Test the available implementation with the same peer, traffic pattern, and qdisc. Include loaded latency and competing traffic in the result. Record kernel and tool versions so that the comparison can be repeated after an update.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Should we disable GRO and TSO?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Do not disable them as a general 10Gbps tuning step. These features reduce or move packet-processing work. Test a specific feature only when the workload or a documented issue gives a reason. Record the starting state, inspect dependent features, and compare CPU cost and latency before keeping the change.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Do 10Gbps dedicated servers need jumbo frames?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">No. Jumbo frames are a separate option for compatible paths. They may reduce packet-processing work in a suitable environment, but require end-to-end planning. Confirm endpoints, switches, logical interfaces, and tunnel overhead before testing a larger MTU. Restore the saved value if connectivity or application performance worsens.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Can larger queues make latency worse?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Yes. Extra queue capacity can let packets wait longer under load. A queue that handles short bursts better may still hurt a service with strict response-time limits. Measure application latency during the same traffic test. If extra depth only postpones drops while processing remains overloaded, investigate processing capacity or traffic demand.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">How do we undo Linux network tuning changes?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Restore the exact saved value for each changed setting, inspect the live result, and check normal traffic. Then remove or correct the persistent file or profile that could reapply the trial. For qdiscs, use a prepared reconstruction procedure. For channel changes, recheck IRQ identities before restoring placement.<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Linux network tuning for 10Gbps starts with throughput, latency, packet-drop, and CPU measurements. A useful change addresses a measured limit, [&hellip;]<\/p>\n","protected":false},"author":2,"featured_media":25657,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"default","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[1],"tags":[],"class_list":["post-25656","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-enterprise-grade-server"],"acf":[],"_links":{"self":[{"href":"https:\/\/atalnetworks.com\/fr\/wp-json\/wp\/v2\/posts\/25656","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/atalnetworks.com\/fr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/atalnetworks.com\/fr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/atalnetworks.com\/fr\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/atalnetworks.com\/fr\/wp-json\/wp\/v2\/comments?post=25656"}],"version-history":[{"count":2,"href":"https:\/\/atalnetworks.com\/fr\/wp-json\/wp\/v2\/posts\/25656\/revisions"}],"predecessor-version":[{"id":25659,"href":"https:\/\/atalnetworks.com\/fr\/wp-json\/wp\/v2\/posts\/25656\/revisions\/25659"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/atalnetworks.com\/fr\/wp-json\/wp\/v2\/media\/25657"}],"wp:attachment":[{"href":"https:\/\/atalnetworks.com\/fr\/wp-json\/wp\/v2\/media?parent=25656"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/atalnetworks.com\/fr\/wp-json\/wp\/v2\/categories?post=25656"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/atalnetworks.com\/fr\/wp-json\/wp\/v2\/tags?post=25656"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}