summaryrefslogtreecommitdiffstats
path: root/include/net/codel.h
Commit message (Collapse)AuthorAgeFilesLines
* Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/netDavid S. Miller2015-05-131-3/+7
|\ | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Four minor merge conflicts: 1) qca_spi.c renamed the local variable used for the SPI device from spi_device to spi, meanwhile the spi_set_drvdata() call got moved further up in the probe function. 2) Two changes were both adding new members to codel params structure, and thus we had overlapping changes to the initializer function. 3) 'net' was making a fix to sk_release_kernel() which is completely removed in 'net-next'. 4) In net_namespace.c, the rtnl_net_fill() call for GET operations had the command value fixed, meanwhile 'net-next' adjusted the argument signature a bit. This also matches example merge resolutions posted by Stephen Rothwell over the past two days. Signed-off-by: David S. Miller <davem@davemloft.net>
| * codel: fix maxpacket/mtu confusionEric Dumazet2015-05-031-3/+7
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Under presence of TSO/GSO/GRO packets, codel at low rates can be quite useless. In following example, not a single packet was ever dropped, while average delay in codel queue is ~100 ms ! qdisc codel 0: parent 1:12 limit 16000p target 5.0ms interval 100.0ms Sent 134376498 bytes 88797 pkt (dropped 0, overlimits 0 requeues 0) backlog 13626b 3p requeues 0 count 0 lastcount 0 ldelay 96.9ms drop_next 0us maxpacket 9084 ecn_mark 0 drop_overlimit 0 This comes from a confusion of what should be the minimal backlog. It is pretty clear it is not 64KB or whatever max GSO packet ever reached the qdisc. codel intent was to use MTU of the device. After the fix, we finally drop some packets, and rtt/cwnd of my single TCP flow are meeting our expectations. qdisc codel 0: parent 1:12 limit 16000p target 5.0ms interval 100.0ms Sent 102798497 bytes 67912 pkt (dropped 1365, overlimits 0 requeues 0) backlog 6056b 3p requeues 0 count 1 lastcount 1 ldelay 36.3ms drop_next 0us maxpacket 10598 ecn_mark 0 drop_overlimit 0 Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Kathleen Nichols <nichols@pollere.com> Cc: Dave Taht <dave.taht@gmail.com> Cc: Van Jacobson <vanj@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
* | codel: add ce_threshold attributeEric Dumazet2015-05-101-1/+11
|/ | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | For DCTCP or similar ECN based deployments on fabrics with shallow buffers, hosts are responsible for a good part of the buffering. This patch adds an optional ce_threshold to codel & fq_codel qdiscs, so that DCTCP can have feedback from queuing in the host. A DCTCP enabled egress port simply have a queue occupancy threshold above which ECT packets get CE mark. In codel language this translates to a sojourn time, so that one doesn't have to worry about bytes or bandwidth but delays. This makes the host an active participant in the health of the whole network. This also helps experimenting DCTCP in a setup without DCTCP compliant fabric. On following example, ce_threshold is set to 1ms, and we can see from 'ldelay xxx us' that TCP is not trying to go around the 5ms codel target. Queue has more capacity to absorb inelastic bursts (say from UDP traffic), as queues are maintained to an optimal level. lpaa23:~# ./tc -s -d qd sh dev eth1 qdisc mq 1: dev eth1 root Sent 87910654696 bytes 58065331 pkt (dropped 0, overlimits 0 requeues 42961) backlog 3108242b 364p requeues 42961 qdisc codel 8063: dev eth1 parent 1:1 limit 1000p target 5.0ms ce_threshold 1.0ms interval 100.0ms Sent 7363778701 bytes 4863809 pkt (dropped 0, overlimits 0 requeues 5503) rate 2348Mbit 193919pps backlog 255866b 46p requeues 5503 count 0 lastcount 0 ldelay 1.0ms drop_next 0us maxpacket 68130 ecn_mark 0 drop_overlimit 0 ce_mark 72384 qdisc codel 8064: dev eth1 parent 1:2 limit 1000p target 5.0ms ce_threshold 1.0ms interval 100.0ms Sent 7636486190 bytes 5043942 pkt (dropped 0, overlimits 0 requeues 5186) rate 2319Mbit 191538pps backlog 207418b 64p requeues 5186 count 0 lastcount 0 ldelay 694us drop_next 0us maxpacket 68130 ecn_mark 0 drop_overlimit 0 ce_mark 69873 qdisc codel 8065: dev eth1 parent 1:3 limit 1000p target 5.0ms ce_threshold 1.0ms interval 100.0ms Sent 11569360142 bytes 7641602 pkt (dropped 0, overlimits 0 requeues 5554) rate 3041Mbit 251096pps backlog 210446b 59p requeues 5554 count 0 lastcount 0 ldelay 889us drop_next 0us maxpacket 68130 ecn_mark 0 drop_overlimit 0 ce_mark 37780 ... Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Florian Westphal <fw@strlen.de> Cc: Daniel Borkmann <daniel@iogearbox.net> Cc: Glenn Judd <glenn.judd@morganstanley.com> Cc: Nandita Dukkipati <nanditad@google.com> Cc: Neal Cardwell <ncardwell@google.com> Cc: Yuchung Cheng <ycheng@google.com> Acked-by: Neal Cardwell <ncardwell@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
* net: use ktime_get_ns() and ktime_get_real_ns() helpersEric Dumazet2014-08-221-1/+1
| | | | | | | | | ktime_get_ns() replaces ktime_to_ns(ktime_get()) ktime_get_real_ns() replaces ktime_to_ns(ktime_get_real()) Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
* net: introduce reciprocal_scale helper and convert usersDaniel Borkmann2014-01-211-3/+1
| | | | | | | | | | | | | | | | | | | As David Laight suggests, we shouldn't necessarily call this reciprocal_divide() when users didn't requested a reciprocal_value(); lets keep the basic idea and call it reciprocal_scale(). More background information on this topic can be found in [1]. Joint work with Hannes Frederic Sowa. [1] http://homepage.cs.uiowa.edu/~jones/bcd/divide.html Suggested-by: David Laight <david.laight@aculab.com> Cc: Jakub Zawadzki <darkjames-ws@darkjames.pl> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: linux-kernel@vger.kernel.org Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
* net: codel: Avoid undefined behavior from signed overflowJesper Dangaard Brouer2013-11-041-4/+15
| | | | | | | | | | | | | | | | | | | | | | As described in commit 5a581b367 (jiffies: Avoid undefined behavior from signed overflow), according to the C standard 3.4.3p3, overflow of a signed integer results in undefined behavior. To fix this, do as the above commit, and do an unsigned subtraction, and interpreting the result as a signed two's-complement number. This is based on the theory from RFC 1982 and is nicely described in wikipedia here: https://en.wikipedia.org/wiki/Serial_number_arithmetic#General_Solution A side-note, I have seen practical issues with the previous logic when dealing with 16-bit, on a 64-bit machine (gcc version 4.4.5). This were 32-bit, which I have not observed issues with. Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Jesper Dangaard Brouer <netoptimizer@brouer.com> Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
* codel: refine one condition to avoid a nul rec_inv_sqrtEric Dumazet2012-08-101-2/+6
| | | | | | | | | | | | | | | | | | | One condition before codel_Newton_step() was not good if we never left the dropping state for a flow. As a result rec_inv_sqrt was 0, instead of the ~0 initial value. codel control law was then set to a very aggressive mode, dropping many packets before reaching 'target' and recovering from this problem. To keep codel_vars_init() as efficient as possible, refine the condition to make sure rec_inv_sqrt initial value is correct Many thanks to Anton Mich for discovering the issue and suggesting a fix. Reported-by: Anton Mich <lp2s1h@gmail.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
* fq_codel: should use qdisc backlog as thresholdEric Dumazet2012-05-161-8/+7
| | | | | | | | | | | | | | | | | | codel_should_drop() logic allows a packet being not dropped if queue size is under max packet size. In fq_codel, we have two possible backlogs : The qdisc global one, and the flow local one. The meaningful one for codel_should_drop() should be the global backlog, not the per flow one, so that thin flows can have a non zero drop/mark probability. Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Dave Taht <dave.taht@bufferbloat.net> Cc: Kathleen Nichols <nichols@pollere.com> Cc: Van Jacobson <van@pollere.net> Signed-off-by: David S. Miller <davem@davemloft.net>
* codel: use u16 field instead of 31bits for rec_inv_sqrtEric Dumazet2012-05-141-10/+15
| | | | | | | | | | | | | David pointed out gcc might generate poor code with 31bit fields. Using u16 is more than enough and permits a better code output. Also make the code intent more readable using constants, fixed point arithmetic not being trivial for everybody. Suggested-by: David Miller <davem@davemloft.net> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
* codel: use Newton method instead of sqrt() and dividesEric Dumazet2012-05-121-31/+37
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | As Van pointed out, interval/sqrt(count) can be implemented using multiplies only. http://en.wikipedia.org/wiki/Methods_of_computing_square_roots#Iterative_methods_for_reciprocal_square_roots This patch implements the Newton method and reciprocal divide. Total cost is 15 cycles instead of 120 on my Corei5 machine (64bit kernel). There is a small 'error' for count values < 5, but we don't really care. I reuse a hole in struct codel_vars : - pack the dropping boolean into one bit - use 31bit to store the reciprocal value of sqrt(count). Suggested-by: Van Jacobson <van@pollere.net> Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Dave Taht <dave.taht@bufferbloat.net> Cc: Kathleen Nichols <nichols@pollere.com> Cc: Tom Herbert <therbert@google.com> Cc: Matt Mathis <mattmathis@google.com> Cc: Yuchung Cheng <ycheng@google.com> Cc: Nandita Dukkipati <nanditad@google.com> Cc: Stephen Hemminger <shemminger@vyatta.com> Signed-off-by: David S. Miller <davem@davemloft.net>
* codel: Controlled Delay AQMEric Dumazet2012-05-101-0/+332
An implementation of CoDel AQM, from Kathleen Nichols and Van Jacobson. http://queue.acm.org/detail.cfm?id=2209336 This AQM main input is no longer queue size in bytes or packets, but the delay packets stay in (FIFO) queue. As we don't have infinite memory, we still can drop packets in enqueue() in case of massive load, but mean of CoDel is to drop packets in dequeue(), using a control law based on two simple parameters : target : target sojourn time (default 5ms) interval : width of moving time window (default 100ms) Based on initial work from Dave Taht. Refactored to help future codel inclusion as a plugin for other linux qdisc (FQ_CODEL, ...), like RED. include/net/codel.h contains codel algorithm as close as possible than Kathleen reference. net/sched/sch_codel.c contains the linux qdisc specific glue. Separate structures permit a memory efficient implementation of fq_codel (to be sent as a separate work) : Each flow has its own struct codel_vars. timestamps are taken at enqueue() time with 1024 ns precision, allowing a range of 2199 seconds in queue, and 100Gb links support. iproute2 uses usec as base unit. Selected packets are dropped, unless ECN is enabled and packets can get ECN mark instead. Tested from 2Mb to 10Gb speeds with no particular problems, on ixgbe and tg3 drivers (BQL enabled). Usage: tc qdisc ... codel [ limit PACKETS ] [ target TIME ] [ interval TIME ] [ ecn ] qdisc codel 10: parent 1:1 limit 2000p target 3.0ms interval 60.0ms ecn Sent 13347099587 bytes 8815805 pkt (dropped 0, overlimits 0 requeues 0) rate 202365Kbit 16708pps backlog 113550b 75p requeues 0 count 116 lastcount 98 ldelay 4.3ms dropping drop_next 816us maxpacket 1514 ecn_mark 84399 drop_overlimit 0 CoDel must be seen as a base module, and should be used keeping in mind there is still a FIFO queue. So a typical setup will probably need a hierarchy of several qdiscs and packet classifiers to be able to meet whatever constraints a user might have. One possible example would be to use fq_codel, which combines Fair Queueing and CoDel, in replacement of sfq / sfq_red. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Dave Taht <dave.taht@bufferbloat.net> Cc: Kathleen Nichols <nichols@pollere.com> Cc: Van Jacobson <van@pollere.net> Cc: Tom Herbert <therbert@google.com> Cc: Matt Mathis <mattmathis@google.com> Cc: Yuchung Cheng <ycheng@google.com> Cc: Stephen Hemminger <shemminger@vyatta.com> Signed-off-by: David S. Miller <davem@davemloft.net>
OpenPOWER on IntegriCloud