{ "content": [ { "type": "text", "text": "# SFB (info)\n\n## NAME\n\nsfb - Stochastic Fair Blue\n\n## SYNOPSIS\n\ntc qdisc ... blue rehash milliseconds db milliseconds limit packets max\npackets target packets increment float decrement float penaltyrate\npackets per second penaltyburst packets\n\n## DESCRIPTION\n\nStochastic Fair Blue is a classless qdisc to manage congestion based on\npacket loss and link utilization history while trying to prevent non-\nresponsive flows (i.e. flows that do not react to congestion marking or\ndropped packets) from impacting performance of responsive flows. Un-\nlike RED, where the marking probability has to be configured, BLUE\ntries to determine the ideal marking probability automatically.\n\n## Sections\n\n- **NAME**\n- **SYNOPSIS**\n- **DESCRIPTION**\n- **ALGORITHM**\n- **LIMITATIONS**\n- **PARAMETERS**\n- **STATISTICS**\n- **NOTES**\n- **SEE ALSO**\n- **SOURCES**\n- **AUTHORS**\n\nUse structuredContent.sections for detailed options, examples, and full documentation.\n" } ], "structuredContent": { "command": "SFB", "section": "", "mode": "info", "summary": "sfb - Stochastic Fair Blue", "synopsis": "tc qdisc ... blue rehash milliseconds db milliseconds limit packets max\npackets target packets increment float decrement float penaltyrate\npackets per second penaltyburst packets", "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" }, { "name": "tc-sfq", "section": "8", "url": "https://www.chedong.com/phpMan.php/man/tc-sfq/8/json" } ], "section_outline": [ { "name": "NAME", "lines": 2, "subsections": [] }, { "name": "SYNOPSIS", "lines": 4, "subsections": [] }, { "name": "DESCRIPTION", "lines": 7, "subsections": [] }, { "name": "ALGORITHM", "lines": 22, "subsections": [] }, { "name": "LIMITATIONS", "lines": 16, "subsections": [] }, { "name": "PARAMETERS", "lines": 55, "subsections": [] }, { "name": "STATISTICS", "lines": 31, "subsections": [] }, { "name": "NOTES", "lines": 20, "subsections": [] }, { "name": "SEE ALSO", "lines": 2, "subsections": [] }, { "name": "SOURCES", "lines": 4, "subsections": [] }, { "name": "AUTHORS", "lines": 4, "subsections": [] } ], "sections": { "NAME": { "content": "sfb - Stochastic Fair Blue\n", "subsections": [] }, "SYNOPSIS": { "content": "tc qdisc ... blue rehash milliseconds db milliseconds limit packets max\npackets target packets increment float decrement float penaltyrate\npackets per second penaltyburst packets\n", "subsections": [] }, "DESCRIPTION": { "content": "Stochastic Fair Blue is a classless qdisc to manage congestion based on\npacket loss and link utilization history while trying to prevent non-\nresponsive flows (i.e. flows that do not react to congestion marking or\ndropped packets) from impacting performance of responsive flows. Un-\nlike RED, where the marking probability has to be configured, BLUE\ntries to determine the ideal marking probability automatically.\n", "subsections": [] }, "ALGORITHM": { "content": "The BLUE algorithm maintains a probability which is used to mark or\ndrop packets that are to be queued. If the queue overflows, the\nmark/drop probability is increased. If the queue becomes empty, the\nprobability is decreased. The Stochastic Fair Blue (SFB) algorithm is\ndesigned to protect TCP flows against non-responsive flows.\n\nThis SFB implementation maintains 8 levels of 16 bins each for account-\ning. Each flow is mapped into a bin of each level using a per-level\nhash value.\n\nEvery bin maintains a marking probability, which gets increased or de-\ncreased based on bin occupancy. If the number of packets exceeds the\nsize of that bin, the marking probability is increased. If the number\ndrops to zero, it is decreased.\n\nThe marking probability is based on the minimum value of all bins a\nflow is mapped into, thus, when a flow does not respond to marking or\ngradual packet drops, the marking probability quickly reaches one.\n\nIn this case, the flow is rate-limited to penaltyrate packets per sec-\nond.\n", "subsections": [] }, "LIMITATIONS": { "content": "Due to SFBs nature, it is possible for responsive flows to share all of\nits bins with a non-responsive flow, causing the responsive flow to be\nmisidentified as being non-responsive.\n\nThe probability of a responsive flow to be misidentified is dependent\non the number of non-responsive flows, M. It is (1 - (1 - (1 / 16.0))\nM) 8, so for example with 10 non-responsive flows approximately\n0.2% of responsive flows will be misidentified.\n\nTo mitigate this, SFB performs performs periodic re-hashing to avoid\nmisclassification for prolonged periods of time.\n\nThe default hashing method will use source and destination ip addresses\nand port numbers if possible, and also supports tunneling protocols.\nAlternatively, an external classifier can be configured, too.\n", "subsections": [] }, "PARAMETERS": { "content": "rehash Time interval in milliseconds when queue perturbation occurs to\navoid erroneously detecting unrelated, responsive flows as being\npart of a non-responsive flow for prolonged periods of time.\nDefaults to 10 minutes.\n\ndb Double buffering warmup wait time, in milliseconds. To avoid\ndestroying the probability history when rehashing is performed,\nthis implementation maintains a second set of levels/bins as de-\nscribed in section 4.4 of the SFB reference. While one set is\nused to manage the queue, a second set is warmed up: Whenever a\nflow is then determined to be non-responsive, the marking proba-\nbilities in the second set are updated. When the rehashing hap-\npens, these bins will be used to manage the queue and all non-\nresponsive flows can be rate-limited immediately. This value\ndetermines how much time has to pass before the 2nd set will\nstart to be warmed up. Defaults to one minute, should be lower\nthan rehash.\n\nlimit Hard limit on the real (not average) total queue size in pack-\nets. Further packets are dropped. Defaults to the transmit\nqueue length of the device the qdisc is attached to.\n\nmax Maximum length of a buckets queue, in packets, before packets\nstart being dropped. Should be slightly larger than target , but\nshould not be set to values exceeding 1.5 times that of target .\nDefaults to 25.\n\ntarget The desired average bin length. If the bin queue length reaches\nthis value, the marking probability is increased by increment.\nThe default value depends on the max setting, with max set to 25\ntarget will default to 20.\n\nincrement\nA value used to increase the marking probability when the queue\nappears to be over-used. Must be between 0 and 1.0. Defaults to\n0.00050.\n\ndecrement\nValue used to decrease the marking probability when the queue is\nfound to be empty. Must be between 0 and 1.0. Defaults to\n0.00005.\n\npenaltyrate\nThe maximum number of packets belonging to flows identified as\nbeing non-responsive that can be enqueued per second. Once this\nnumber has been reached, further packets of such non-responsive\nflows are dropped. Set this to a reasonable fraction of your\nuplink throughput; the default value of 10 packets is probably\ntoo small.\n\npenaltyburst\nThe number of packets a flow is permitted to exceed the penalty\nrate before packets start being dropped. Defaults to 20 pack-\nets.\n", "subsections": [] }, "STATISTICS": { "content": "This qdisc exposes additional statistics via 'tc -s qdisc' output.\nThese are:\n\nearlydrop\nThe number of packets dropped before a per-flow queue was full.\n\nratedrop\nThe number of packets dropped because of rate-limiting. If this\nvalue is high, there are many non-reactive flows being sent\nthrough sfb. In such cases, it might be better to embed sfb\nwithin a classful qdisc to better control such flows using a\ndifferent, shaping qdisc.\n\nbucketdrop\nThe number of packets dropped because a per-flow queue was full.\nHigh bucketdrop may point to a high number of aggressive, short-\nlived flows.\n\nqueuedrop\nThe number of packets dropped due to reaching limit. This should\nnormally be 0.\n\nmarked The number of packets marked with ECN.\n\nmaxqlen\nThe length of the current longest per-flow (virtual) queue.\n\nmaxprob\nThe maximum per-flow drop probability. 1 means that some flows\nhave been detected as non-reactive.\n", "subsections": [] }, "NOTES": { "content": "SFB automatically enables use of Explicit Congestion Notification\n(ECN). Also, this SFB implementation does not queue packets itself.\nRather, packets are enqueued to the inner qdisc (defaults to pfifo).\nBecause sfb maintains virtual queue states, the inner qdisc must not\ndrop a packet previously queued. Furthermore, if a buckets queue has a\nvery high marking rate, this implementation will start dropping packets\ninstead of marking them, as such a situation points to either bad con-\ngestion, or an unresponsive flow.\n\nEXAMPLE & USAGE\nTo attach to interface $DEV, using default options:\n\n# tc qdisc add dev $DEV handle 1: root sfb\n\nOnly use destination ip addresses for assigning packets to bins, per-\nturbing hash results every 10 minutes:\n\n# tc filter add dev $DEV parent 1: handle 1 flow hash keys dst perturb\n600\n", "subsections": [] }, "SEE ALSO": { "content": "tc(8), tc-red(8), tc-sfq(8)\n", "subsections": [] }, "SOURCES": { "content": "o W. Feng, D. Kandlur, D. Saha, K. Shin, BLUE: A New Class of Ac-\ntive Queue Management Algorithms, U. Michigan CSE-TR-387-99,\nApril 1999.\n", "subsections": [] }, "AUTHORS": { "content": "This SFB implementation was contributed by Juliusz Chroboczek and Eric\nDumazet.\n\niproute2 August 2011 SFB(8)", "subsections": [] } } } }