tcp: RFC6298 compliant TCP RTO calculation #2

danielmgit · 2016-06-10T14:39:04Z

Linux RTO calculation is adjusted to be RFC6298 Standard compliant.
MinRTO is no longer added to the computed RTO, RTO damping and
overestimation are decreased.

In RFC 6298 Standard TCP Retransmission Timeout (RTO) calculation the
calculated RTO is rounded up to the Minimum RTO (MinRTO), if it is less.
The Linux implementation as a discrepancy to the Standard basically
adds the defined MinRTO to the calculated RTO. When comparing both
approaches, the Linux calculation seems to perform worse for sender
limited TCP flows like Telnet, SSH or constant bit rate encoded
transmissions, especially for Round Trip Times (RTT) of 50ms to 800ms.

Compared to the Linux implementation the RFC 6298 proposed RTO
calculation performs better and more precise in adapting to current
network characteristics. Extensive measurements for bulk data did not
show a negative impact of the adjusted calculation.

Performance Comparison for sender-limited-flows:
Rate: 10Mbit/s, Delay: 200ms, Delay Variation: 10ms, Time between each
scheduled segment: 1s, Amount Data Segments: 300, Mean of 8 runs

Mean Response Waiting Time [milliseconds]

old 205.8 208.3 217.0 220.3 227.8 249.9 271.0 308.9
new 204.3 206.5 207.1 210.5 217.3 224.2 237.8 258.3

0.5    1      1.5    2      3      5      7      10

Packet Error Rate [percent]

Detailed Analysis:
https://docs.google.com/document/d/1pKmPfnQb6fDK4qpiNVkN8cQyGE4wYDZukcuZfR-BnnM/edit?usp=sharing

Signed-off-by: Daniel Metz dmetz@mytum.de

Linux RTO calculation is adjusted to be RFC6298 Standard compliant. MinRTO is no longer added to the computed RTO, RTO damping and overestimation are decreased. In RFC 6298 Standard TCP Retransmission Timeout (RTO) calculation the calculated RTO is rounded up to the Minimum RTO (MinRTO), if it is less. The Linux implementation as a discrepancy to the Standard basically adds the defined MinRTO to the calculated RTO. When comparing both approaches, the Linux calculation seems to perform worse for sender limited TCP flows like Telnet, SSH or constant bit rate encoded transmissions, especially for Round Trip Times (RTT) of 50ms to 800ms. Compared to the Linux implementation the RFC 6298 proposed RTO calculation performs better and more precise in adapting to current network characteristics. Extensive measurements for bulk data did not show a negative impact of the adjusted calculation. Performance Comparison for sender-limited-flows: Rate: 10Mbit/s, Delay: 200ms, Delay Variation: 10ms, Time between each scheduled segment: 1s, Amount Data Segments: 300, Mean of 8 runs Mean Response Waiting Time [milliseconds] old 205.8 208.3 217.0 220.3 227.8 249.9 271.0 308.9 new 204.3 206.5 207.1 210.5 217.3 224.2 237.8 258.3 0.5 1 1.5 2 3 5 7 10 Packet Error Rate [percent] Detailed Analysis: https://docs.google.com/document/d/1pKmPfnQb6fDK4qpiNVkN8cQyGE4wYDZukcuZfR-BnnM/edit?usp=sharing Signed-off-by: Daniel Metz <dmetz@mytum.de>

Zi Shen Lim says: ==================== arm64 BPF JIT updates Updates for arm64 eBPF JIT. The main addition here is implementation of bpf_tail_call. Changes since v2: - None. Resubmit per David Miller. Changes since v1: - Added patch #1 to address build error due to missing header inclusion in linux/bpf.h. (Thanks to suggestion and ack by Daniel Borkmann) Ordered it ahead of bpf_tail_call patch #2 so build error is not triggered. ==================== Signed-off-by: David S. Miller <davem@davemloft.net>

With zpci_disable() working, lockdep detected a potential deadlock (lockdep output at the end). The deadlock is between recovering a PCI function via the /sys/bus/pci/devices/<dev>/recover attribute vs powering it off via /sys/bus/pci/slots/<slot>/power. The fix is analogous to the changes in commit 0ee223b ("scsi: core: Avoid that SCSI device removal through sysfs triggers a deadlock") that fixed a potential deadlock on removing a SCSI device via sysfs. [ 204.830107] ====================================================== [ 204.830109] WARNING: possible circular locking dependency detected [ 204.830111] 5.5.0-rc2-06072-gbc03ecc9a672 #6 Tainted: G W [ 204.830112] ------------------------------------------------------ [ 204.830113] bash/1034 is trying to acquire lock: [ 204.830115] 0000000192a1a610 (kn->count#200){++++}, at: kernfs_remove_by_name_ns+0x5c/0xa8 [ 204.830122] but task is already holding lock: [ 204.830123] 00000000c16134a8 (pci_rescan_remove_lock){+.+.}, at: pci_stop_and_remove_bus_device_locked+0x26/0x48 [ 204.830128] which lock already depends on the new lock. [ 204.830129] the existing dependency chain (in reverse order) is: [ 204.830130] -> #1 (pci_rescan_remove_lock){+.+.}: [ 204.830134] validate_chain+0x93a/0xd08 [ 204.830136] __lock_acquire+0x4ae/0x9d0 [ 204.830137] lock_acquire+0x114/0x280 [ 204.830140] __mutex_lock+0xa2/0x960 [ 204.830142] mutex_lock_nested+0x32/0x40 [ 204.830145] recover_store+0x4c/0xa8 [ 204.830147] kernfs_fop_write+0xe6/0x218 [ 204.830151] vfs_write+0xb0/0x1b8 [ 204.830152] ksys_write+0x6c/0xf8 [ 204.830154] system_call+0xd8/0x2d8 [ 204.830155] -> #0 (kn->count#200){++++}: [ 204.830187] check_noncircular+0x1e6/0x240 [ 204.830189] check_prev_add+0xfc/0xdb0 [ 204.830190] validate_chain+0x93a/0xd08 [ 204.830192] __lock_acquire+0x4ae/0x9d0 [ 204.830193] lock_acquire+0x114/0x280 [ 204.830194] __kernfs_remove.part.0+0x2e4/0x360 [ 204.830196] kernfs_remove_by_name_ns+0x5c/0xa8 [ 204.830198] remove_files.isra.0+0x4c/0x98 [ 204.830199] sysfs_remove_group+0x66/0xc8 [ 204.830201] sysfs_remove_groups+0x46/0x68 [ 204.830204] device_remove_attrs+0x52/0x90 [ 204.830207] device_del+0x182/0x418 [ 204.830208] pci_remove_bus_device+0x8a/0x130 [ 204.830210] pci_stop_and_remove_bus_device_locked+0x3a/0x48 [ 204.830212] disable_slot+0x68/0x100 [ 204.830213] power_write_file+0x7c/0x130 [ 204.830215] kernfs_fop_write+0xe6/0x218 [ 204.830217] vfs_write+0xb0/0x1b8 [ 204.830218] ksys_write+0x6c/0xf8 [ 204.830220] system_call+0xd8/0x2d8 [ 204.830221] other info that might help us debug this: [ 204.830223] Possible unsafe locking scenario: [ 204.830224] CPU0 CPU1 [ 204.830225] ---- ---- [ 204.830226] lock(pci_rescan_remove_lock); [ 204.830227] lock(kn->count#200); [ 204.830229] lock(pci_rescan_remove_lock); [ 204.830231] lock(kn->count#200); [ 204.830233] *** DEADLOCK *** [ 204.830234] 4 locks held by bash/1034: [ 204.830235] #0: 00000001b6fbc498 (sb_writers#4){.+.+}, at: vfs_write+0x158/0x1b8 [ 204.830239] #1: 000000018c9f5090 (&of->mutex){+.+.}, at: kernfs_fop_write+0xaa/0x218 [ 204.830242] #2: 00000001f7da0810 (kn->count#235){.+.+}, at: kernfs_fop_write+0xb6/0x218 [ 204.830245] #3: 00000000c16134a8 (pci_rescan_remove_lock){+.+.}, at: pci_stop_and_remove_bus_device_locked+0x26/0x48 [ 204.830248] stack backtrace: [ 204.830250] CPU: 2 PID: 1034 Comm: bash Tainted: G W 5.5.0-rc2-06072-gbc03ecc9a672 #6 [ 204.830252] Hardware name: IBM 8561 T01 703 (LPAR) [ 204.830253] Call Trace: [ 204.830257] [<00000000c05e10c0>] show_stack+0x88/0xf0 [ 204.830260] [<00000000c112dca4>] dump_stack+0xa4/0xe0 [ 204.830261] [<00000000c0694c06>] check_noncircular+0x1e6/0x240 [ 204.830263] [<00000000c0695bec>] check_prev_add+0xfc/0xdb0 [ 204.830264] [<00000000c06971da>] validate_chain+0x93a/0xd08 [ 204.830266] [<00000000c06994c6>] __lock_acquire+0x4ae/0x9d0 [ 204.830267] [<00000000c069867c>] lock_acquire+0x114/0x280 [ 204.830269] [<00000000c09ca15c>] __kernfs_remove.part.0+0x2e4/0x360 [ 204.830270] [<00000000c09cb5c4>] kernfs_remove_by_name_ns+0x5c/0xa8 [ 204.830272] [<00000000c09cee14>] remove_files.isra.0+0x4c/0x98 [ 204.830274] [<00000000c09cf2ae>] sysfs_remove_group+0x66/0xc8 [ 204.830276] [<00000000c09cf356>] sysfs_remove_groups+0x46/0x68 [ 204.830278] [<00000000c0e3dfe2>] device_remove_attrs+0x52/0x90 [ 204.830280] [<00000000c0e40382>] device_del+0x182/0x418 [ 204.830281] [<00000000c0dcfd7a>] pci_remove_bus_device+0x8a/0x130 [ 204.830283] [<00000000c0dcfe92>] pci_stop_and_remove_bus_device_locked+0x3a/0x48 [ 204.830285] [<00000000c0de7190>] disable_slot+0x68/0x100 [ 204.830286] [<00000000c0de6514>] power_write_file+0x7c/0x130 [ 204.830288] [<00000000c09cc846>] kernfs_fop_write+0xe6/0x218 [ 204.830290] [<00000000c08f3480>] vfs_write+0xb0/0x1b8 [ 204.830291] [<00000000c08f378c>] ksys_write+0x6c/0xf8 [ 204.830293] [<00000000c1154374>] system_call+0xd8/0x2d8 [ 204.830294] INFO: lockdep is turned off. Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com> Reviewed-by: Peter Oberparleiter <oberpar@linux.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

Hayes Wang says: ==================== r8152: serial fixes v3: 1. Fix the typos for patch #5 and #6. 2. Modify the commit message of patch torvalds#9. v2: For patch #2, move declaring the variable "ocp_data". v1: These patches are used to fix some issues for RTL8153. ==================== Signed-off-by: David S. Miller <davem@davemloft.net>

Gautam Ramakrishnan says: ==================== net: sched: add Flow Queue PIE packet scheduler Flow Queue PIE packet scheduler This patch series implements the Flow Queue Proportional Integral controller Enhanced (FQ-PIE) active queue Management algorithm. It is an enhancement over the PIE algorithm. It integrates the PIE aqm with a deficit round robin scheme. FQ-PIE is implemented over the latest version of PIE which uses timestamps to calculate queue delay with an additional option of using average dequeue rate to calculate the queue delay. This patch also adds a memory limit of all the packets across all queues to a default value of 32Mb. - Patch #1 - Creates pie.h and moves all small functions and structures common to PIE and FQ-PIE here. The functions are all made inline. - Patch #2 - #8 - Addresses code formatting, indentation, comment changes and rearrangement of structure members. - Patch torvalds#9 - Refactors sch_pie.c by changing arguments to calculate_probability(), [pie_]drop_early() and pie_process_dequeue() to make it generic enough to be used by sch_fq_pie.c. These functions are exported to be used by sch_fq_pie.c. - Patch torvalds#10 - Adds the FQ-PIE Qdisc. For more information: https://tools.ietf.org/html/rfc8033 Changes from v6 to v7 - Call tcf_block_put() when destroying the Qdisc as suggested by Jakub Kicinski. Changes from v5 to v6 - Rearranged struct members according to their access pattern and to remove holes. Changes from v4 to v5 - This patch series breaks down patch 1 of v4 into separate logical commits as suggested by David Miller. Changes from v3 to v4 - Used non deprecated version of nla_parse_nested - Used SZ_32M macro - Removed an unused variable - Code cleanup All suggested by Jakub and Toke. Changes from v2 to v3 - Exported drop_early, pie_process_dequeue and calculate_probability functions from sch_pie as suggested by Stephen Hemminger. Changes from v1 ( and RFC patch) to v2 - Added timestamp to calculate queue delay as recommended by Dave Taht - Packet memory limit implemented as recommended by Toke. - Added external classifier as recommended by Toke. - Used NET_XMIT_CN instead of NET_XMIT_DROP as the return value in the fq_pie_qdisc_enqueue function. ==================== Signed-off-by: David S. Miller <davem@davemloft.net>

We don't need to hold the local pinctrl lock here to set irq wake on the summary irq line. Doing so only leads to lockdep warnings instead of protecting us from anything. Remove the locking. WARNING: possible circular locking dependency detected 5.4.11 #2 Tainted: G W ------------------------------------------------------ cat/3083 is trying to acquire lock: ffffff81f4fa58c0 (&irq_desc_lock_class){-.-.}, at: __irq_get_desc_lock+0x64/0x94 but task is already holding lock: ffffff81f4880c18 (&pctrl->lock){-.-.}, at: msm_gpio_irq_set_wake+0x48/0x7c which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #1 (&pctrl->lock){-.-.}: _raw_spin_lock_irqsave+0x64/0x80 msm_gpio_irq_ack+0x68/0xf4 __irq_do_set_handler+0xe0/0x180 __irq_set_handler+0x60/0x9c irq_domain_set_info+0x90/0xb4 gpiochip_hierarchy_irq_domain_alloc+0x110/0x200 __irq_domain_alloc_irqs+0x130/0x29c irq_create_fwspec_mapping+0x1f0/0x300 irq_create_of_mapping+0x70/0x98 of_irq_get+0xa4/0xd4 spi_drv_probe+0x4c/0xb0 really_probe+0x138/0x3f0 driver_probe_device+0x70/0x140 __device_attach_driver+0x9c/0x110 bus_for_each_drv+0x88/0xd0 __device_attach+0xb0/0x160 device_initial_probe+0x20/0x2c bus_probe_device+0x34/0x94 device_add+0x35c/0x3f0 spi_add_device+0xbc/0x194 of_register_spi_devices+0x2c8/0x408 spi_register_controller+0x57c/0x6fc spi_geni_probe+0x260/0x328 platform_drv_probe+0x90/0xb0 really_probe+0x138/0x3f0 driver_probe_device+0x70/0x140 device_driver_attach+0x4c/0x6c __driver_attach+0xcc/0x154 bus_for_each_dev+0x84/0xcc driver_attach+0x2c/0x38 bus_add_driver+0x108/0x1fc driver_register+0x64/0xf8 __platform_driver_register+0x4c/0x58 spi_geni_driver_init+0x1c/0x24 do_one_initcall+0x1a4/0x3e8 do_initcall_level+0xb4/0xcc do_basic_setup+0x30/0x48 kernel_init_freeable+0x124/0x1a8 kernel_init+0x14/0x100 ret_from_fork+0x10/0x18 -> #0 (&irq_desc_lock_class){-.-.}: __lock_acquire+0xeb4/0x2388 lock_acquire+0x1cc/0x210 _raw_spin_lock_irqsave+0x64/0x80 __irq_get_desc_lock+0x64/0x94 irq_set_irq_wake+0x40/0x144 msm_gpio_irq_set_wake+0x5c/0x7c set_irq_wake_real+0x40/0x5c irq_set_irq_wake+0x70/0x144 cros_ec_rtc_suspend+0x38/0x4c platform_pm_suspend+0x34/0x60 dpm_run_callback+0x64/0xcc __device_suspend+0x310/0x41c dpm_suspend+0xf8/0x298 dpm_suspend_start+0x84/0xb4 suspend_devices_and_enter+0xbc/0x620 pm_suspend+0x210/0x348 state_store+0xb0/0x108 kobj_attr_store+0x14/0x24 sysfs_kf_write+0x4c/0x64 kernfs_fop_write+0x15c/0x1fc __vfs_write+0x54/0x18c vfs_write+0xe4/0x1a4 ksys_write+0x7c/0xe4 __arm64_sys_write+0x20/0x2c el0_svc_common+0xa8/0x160 el0_svc_handler+0x7c/0x98 el0_svc+0x8/0xc other info that might help us debug this: Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(&pctrl->lock); lock(&irq_desc_lock_class); lock(&pctrl->lock); lock(&irq_desc_lock_class); *** DEADLOCK *** 7 locks held by cat/3083: #0: ffffff81f06d1420 (sb_writers#7){.+.+}, at: vfs_write+0xd0/0x1a4 #1: ffffff81c8935680 (&of->mutex){+.+.}, at: kernfs_fop_write+0x12c/0x1fc #2: ffffff81f4c322f0 (kn->count#337){.+.+}, at: kernfs_fop_write+0x134/0x1fc #3: ffffffe89a641d60 (system_transition_mutex){+.+.}, at: pm_suspend+0x108/0x348 #4: ffffff81f190e970 (&dev->mutex){....}, at: __device_suspend+0x168/0x41c #5: ffffff81f183d8c0 (lock_class){-.-.}, at: __irq_get_desc_lock+0x64/0x94 #6: ffffff81f4880c18 (&pctrl->lock){-.-.}, at: msm_gpio_irq_set_wake+0x48/0x7c stack backtrace: CPU: 4 PID: 3083 Comm: cat Tainted: G W 5.4.11 #2 Hardware name: Google Cheza (rev3+) (DT) Call trace: dump_backtrace+0x0/0x174 show_stack+0x20/0x2c dump_stack+0xc8/0x124 print_circular_bug+0x2ac/0x2c4 check_noncircular+0x1a0/0x1a8 __lock_acquire+0xeb4/0x2388 lock_acquire+0x1cc/0x210 _raw_spin_lock_irqsave+0x64/0x80 __irq_get_desc_lock+0x64/0x94 irq_set_irq_wake+0x40/0x144 msm_gpio_irq_set_wake+0x5c/0x7c set_irq_wake_real+0x40/0x5c irq_set_irq_wake+0x70/0x144 cros_ec_rtc_suspend+0x38/0x4c platform_pm_suspend+0x34/0x60 dpm_run_callback+0x64/0xcc __device_suspend+0x310/0x41c dpm_suspend+0xf8/0x298 dpm_suspend_start+0x84/0xb4 suspend_devices_and_enter+0xbc/0x620 pm_suspend+0x210/0x348 state_store+0xb0/0x108 kobj_attr_store+0x14/0x24 sysfs_kf_write+0x4c/0x64 kernfs_fop_write+0x15c/0x1fc __vfs_write+0x54/0x18c vfs_write+0xe4/0x1a4 ksys_write+0x7c/0xe4 __arm64_sys_write+0x20/0x2c el0_svc_common+0xa8/0x160 el0_svc_handler+0x7c/0x98 el0_svc+0x8/0xc Fixes: 6aced33 ("pinctrl: msm: drop wake_irqs bitmap") Cc: Douglas Anderson <dianders@chromium.org> Cc: Brian Masney <masneyb@onstation.org> Cc: Lina Iyer <ilina@codeaurora.org> Cc: Maulik Shah <mkshah@codeaurora.org> Signed-off-by: Stephen Boyd <swboyd@chromium.org> Link: https://lore.kernel.org/r/20200121180950.36959-1-swboyd@chromium.org Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org> Signed-off-by: Linus Walleij <linus.walleij@linaro.org>

@dst

With the introduction of MOVDIR64B instruction, there is now an instruction that can write 64 bytes of data atomically. Quoting from Intel SDM: "There is no atomicity guarantee provided for the 64-byte load operation from source address, and processor implementations may use multiple load operations to read the 64-bytes. The 64-byte direct-store issued by MOVDIR64B guarantees 64-byte write-completion atomicity. This means that the data arrives at the destination in a single undivided 64-byte write transaction." We have identified at least 3 different use cases for this instruction in the format of func(dst, src, count): 1) Clear poison / Initialize MKTME memory @dst is normal memory. @src in normal memory. Does not increment. (Copy same line to all targets) @count (to clear/init multiple lines) 2) Submit command(s) to new devices @dst is a special MMIO region for a device. Does not increment. @src is normal memory. Increments. @count usually is 1, but can be multiple. 3) Copy to iomem in big chunks @dst is iomem and increments @src in normal memory and increments @count is number of chunks to copy Add support for case #2 to support device that will accept commands via this instruction. We provide a @count in order to submit a batch of preprogrammed descriptors in virtually contiguous memory. This allows the caller to submit multiple descriptors to a device with a single submission. The special device requires the entire 64bytes descriptor to be written atomically and will accept MOVDIR64B instruction. Signed-off-by: Dave Jiang <dave.jiang@intel.com> Acked-by: Borislav Petkov <bp@suse.de> Link: https://lore.kernel.org/r/157965022175.73301.10174614665472962675.stgit@djiang5-desk3.ch.intel.com Signed-off-by: Vinod Koul <vkoul@kernel.org>

Ido Schimmel says: ==================== mlxsw: Offload TBF Petr says: In order to allow configuration of shapers on Spectrum family of machines, recognize TBF either as root Qdisc, or as a child of ETS or PRIO. Configure rate of maximum shaper according to TBF rate setting, and maximum shaper burst size according to TBF burst setting. - Patches #1 and #2 make the TBF shaper suitable for offloading. - Patches #3, #4 and #5 are refactoring aimed at easier support of leaf Qdiscs in general. - Patches #6 to torvalds#10 gradually introduce TBF offload. - Patches torvalds#11 to torvalds#14 add selftests. ==================== Signed-off-by: David S. Miller <davem@davemloft.net>

danielmgit closed this Jun 13, 2016

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

tcp: RFC6298 compliant TCP RTO calculation #2

tcp: RFC6298 compliant TCP RTO calculation #2

danielmgit commented Jun 10, 2016

tcp: RFC6298 compliant TCP RTO calculation #2

tcp: RFC6298 compliant TCP RTO calculation #2

Conversation

danielmgit commented Jun 10, 2016