{ "content": [ { "type": "text", "text": "# tc-sfq (info)\n\n## NAME\n\nsfq - Stochastic Fairness Queueing\n\n## SYNOPSIS\n\ntc qdisc ... [ divisor hashtablesize ] [ limit packets ] [ perturb\nseconds ] [ quantum bytes ] [ flows number ] [ depth number ] [ head-\ndrop ] [ redflowlimit bytes ] [ min bytes ] [ max bytes ] [ avpkt bytes\n] [ burst packets ] [ probability P ] [ ecn ] [ harddrop ]\n\n## DESCRIPTION\n\nStochastic Fairness Queueing is a classless queueing discipline avail-\nable for traffic control with the tc(8) command.\n\n## Sections\n\n- **NAME**\n- **SYNOPSIS**\n- **DESCRIPTION**\n- **ALGORITHM**\n- **PARAMETERS**\n- **SOURCE**\n- **SEE ALSO**\n- **AUTHORS**\n\nUse structuredContent.sections for detailed options, examples, and full documentation.\n" } ], "structuredContent": { "command": "tc-sfq", "section": "", "mode": "info", "summary": "sfq - Stochastic Fairness Queueing", "synopsis": "tc qdisc ... [ divisor hashtablesize ] [ limit packets ] [ perturb\nseconds ] [ quantum bytes ] [ flows number ] [ depth number ] [ head-\ndrop ] [ redflowlimit bytes ] [ min bytes ] [ max bytes ] [ avpkt bytes\n] [ burst packets ] [ probability P ] [ ecn ] [ harddrop ]", "tldr_summary": null, "tldr_examples": [], "tldr_source": null, "flags": [], "examples": [], "see_also": [ { "name": "tc", "section": "8", "url": "https://www.chedong.com/phpMan.php/man/tc/8/json" }, { "name": "tc-red", "section": "8", "url": "https://www.chedong.com/phpMan.php/man/tc-red/8/json" } ], "section_outline": [ { "name": "NAME", "lines": 2, "subsections": [] }, { "name": "SYNOPSIS", "lines": 5, "subsections": [] }, { "name": "DESCRIPTION", "lines": 14, "subsections": [] }, { "name": "ALGORITHM", "lines": 38, "subsections": [] }, { "name": "PARAMETERS", "lines": 103, "subsections": [] }, { "name": "SOURCE", "lines": 9, "subsections": [] }, { "name": "SEE ALSO", "lines": 2, "subsections": [] }, { "name": "AUTHORS", "lines": 6, "subsections": [] } ], "sections": { "NAME": { "content": "sfq - Stochastic Fairness Queueing\n", "subsections": [] }, "SYNOPSIS": { "content": "tc qdisc ... [ divisor hashtablesize ] [ limit packets ] [ perturb\nseconds ] [ quantum bytes ] [ flows number ] [ depth number ] [ head-\ndrop ] [ redflowlimit bytes ] [ min bytes ] [ max bytes ] [ avpkt bytes\n] [ burst packets ] [ probability P ] [ ecn ] [ harddrop ]\n", "subsections": [] }, "DESCRIPTION": { "content": "Stochastic Fairness Queueing is a classless queueing discipline avail-\nable for traffic control with the tc(8) command.\n\nSFQ does not shape traffic but only schedules the transmission of pack-\nets, based on 'flows'. The goal is to ensure fairness so that each\nflow is able to send data in turn, thus preventing any single flow from\ndrowning out the rest.\n\nThis may in fact have some effect in mitigating a Denial of Service at-\ntempt.\n\nSFQ is work-conserving and therefore always delivers a packet if it has\none available.\n", "subsections": [] }, "ALGORITHM": { "content": "On enqueueing, each packet is assigned to a hash bucket, based on the\npackets hash value. This hash value is either obtained from an exter-\nnal flow classifier (use tc filter to set them), or a default internal\nclassifier if no external classifier has been configured.\n\nWhen the internal classifier is used, sfq uses\n\n(i) Source address\n\n(ii) Destination address\n\n(iii) Source and Destination port\n\nIf these are available. SFQ knows about ipv4 and ipv6 and also UDP, TCP\nand ESP. Packets with other protocols are hashed based on the 32bits\nrepresentation of their destination and source. A flow corresponds\nmostly to a TCP/IP connection.\n\nEach of these buckets should represent a unique flow. Because multiple\nflows may get hashed to the same bucket, sfqs internal hashing algo-\nrithm may be perturbed at configurable intervals so that the unfairness\nlasts only for a short while. Perturbation may however cause some inad-\nvertent packet reordering to occur. After linux-3.3, there is no packet\nreordering problem, but possible packet drops if rehashing hits one\nlimit (number of flows or packets per flow)\n\nWhen dequeuing, each hashbucket with data is queried in a round robin\nfashion.\n\nBefore linux-3.3, the compile time maximum length of the SFQ is 128\npackets, which can be spread over at most 128 buckets of 1024 avail-\nable. In case of overflow, tail-drop is performed on the fullest\nbucket, thus maintaining fairness.\n\nAfter linux-3.3, maximum length of SFQ is 65535 packets, and divisor\nlimit is 65536. In case of overflow, tail-drop is performed on the\nfullest bucket, unless headdrop was requested.\n", "subsections": [] }, "PARAMETERS": { "content": "divisor\nCan be used to set a different hash table size, available from\nkernel 2.6.39 onwards. The specified divisor must be a power of\ntwo and cannot be larger than 65536. Default value: 1024.\n\nlimit Upper limit of the SFQ. Can be used to reduce the default length\nof 127 packets. After linux-3.3, it can be raised.\n\ndepth Limit of packets per flow (after linux-3.3). Default to 127 and\ncan be lowered.\n\nperturb\nInterval in seconds for queue algorithm perturbation. Defaults\nto 0, which means that no perturbation occurs. Do not set too\nlow for each perturbation may cause some packet reordering or\nlosses. Advised value: 60 This value has no effect when external\nflow classification is used. Its better to increase divisor\nvalue to lower risk of hash collisions.\n\nquantum\nAmount of bytes a flow is allowed to dequeue during a round of\nthe round robin process. Defaults to the MTU of the interface\nwhich is also the advised value and the minimum value.\n\nflows After linux-3.3, it is possible to change the default limit of\nflows. Default value is 127\n\nheaddrop\nDefault SFQ behavior is to perform tail-drop of packets from a\nflow. You can ask a headdrop instead, as this is known to pro-\nvide a better feedback for TCP flows.\n\nredflowlimit\nConfigure the optional RED module on top of each SFQ flow. Ran-\ndom Early Detection principle is to perform packet marks or\ndrops in a probabilistic way. (man tc-red for details about\nRED)\nredflowlimit configures the hard limit on the real (not average) queue size per SFQ flow in bytes.\n\nmin Average queue size at which marking becomes a possibility. De-\nfaults to max /3\n\nmax At this average queue size, the marking probability is maximal.\nDefaults to redflowlimit /4\n\nprobability\nMaximum probability for marking, specified as a floating\npoint number from 0.0 to 1.0. Default value is 0.02\n\navpkt Specified in bytes. Used with burst to determine the time con-\nstant for average queue size calculations. Default value is 1000\n\nburst Used for determining how fast the average queue size is influ-\nenced by the real queue size.\nDefault value is :\n(2 * min + max) / (3 * avpkt)\n\necn RED can either 'mark' or 'drop'. Explicit Congestion Notifica-\ntion allows RED to notify remote hosts that their rate exceeds\nthe amount of bandwidth available. Non-ECN capable hosts can\nonly be notified by dropping a packet. If this parameter is\nspecified, packets which indicate that their hosts honor ECN\nwill only be marked and not dropped, unless the queue size hits\ndepth packets.\n\nharddrop\nIf average flow queue size is above max bytes, this parameter\nforces a drop instead of ecn marking.\n\nEXAMPLE & USAGE\nTo attach to device ppp0:\n\n# tc qdisc add dev ppp0 root sfq\n\nPlease note that SFQ, like all non-shaping (work-conserving) qdiscs, is\nonly useful if it owns the queue. This is the case when the link speed\nequals the actually available bandwidth. This holds for regular phone\nmodems, ISDN connections and direct non-switched ethernet links.\n\nMost often, cable modems and DSL devices do not fall into this cate-\ngory. The same holds for when connected to a switch and trying to send\ndata to a congested segment also connected to the switch.\n\nIn this case, the effective queue does not reside within Linux and is\ntherefore not available for scheduling.\n\nEmbed SFQ in a classful qdisc to make sure it owns the queue.\n\nIt is possible to use external classifiers with sfq, for example to\nhash traffic based only on source/destination ip addresses:\n\n# tc filter add ... flow hash keys src,dst perturb 30 divisor 1024\n\nNote that the given divisor should match the one used by sfq. If you\nhave changed the sfq default of 1024, use the same value for the flow\nhash filter, too.\n\nExample of sfq with optional RED mode :\n\n# tc qdisc add dev eth0 parent 1:1 handle 10: sfq limit 3000 flows 512\ndivisor 16384\nredflowlimit 100000 min 8000 max 60000 probability 0.20 ecn headdrop\n", "subsections": [] }, "SOURCE": { "content": "o Paul E. McKenney \"Stochastic Fairness Queuing\", IEEE INFOCOMM'90\nProceedings, San Francisco, 1990.\n\no Paul E. McKenney \"Stochastic Fairness Queuing\", \"Interworking:\nResearch and Experience\", v.2, 1991, p.113-131.\n\no See also: M. Shreedhar and George Varghese \"Efficient Fair Queu-\ning using Deficit Round Robin\", Proc. SIGCOMM 95.\n", "subsections": [] }, "SEE ALSO": { "content": "tc(8), tc-red(8)\n", "subsections": [] }, "AUTHORS": { "content": "Alexey N. Kuznetsov, , Eric Dumazet .\n\nThis manpage maintained by bert hubert \n\niproute2 24 January 2012 TC(8)", "subsections": [] } } } }