Recommended
More Related Content
What's hot
What's hot (20)
Viewers also liked
Viewers also liked (20)
Similar to Boost UDP Transaction Performance
Similar to Boost UDP Transaction Performance (20)
More from LF Events
More from LF Events (14)
Recently uploaded
Recently uploaded (20)
Boost UDP Transaction Performance
- 1. Copyright © 2016 NTT Corp. All Rights Reserved. Boost UDP Transaction Performance Toshiaki Makita NTT Open Source Software Center
- 2. 2Copyright © 2016 NTT Corp. All Rights Reserved. • Background • Basic technologies for network performance • How to improve UDP performance Today's topics
- 3. 3Copyright © 2016 NTT Corp. All Rights Reserved. • Linux kernel engineer at NTT Open Source Software Center • Technical support for NTT group companies • Active patch submitter on kernel networking subsystem Who is Toshiaki Makita?
- 4. 4Copyright © 2016 NTT Corp. All Rights Reserved. Background
- 5. 5Copyright © 2016 NTT Corp. All Rights Reserved. • Services using UDP • DNS • RADIUS • NTP • SNMP • ... • Heavily used by network service providers UDP transactions in the Internet
- 6. 6Copyright © 2016 NTT Corp. All Rights Reserved. • Ethernet bandwidth evolution • 10M -> 100M -> 1G -> 10G -> 40G -> 100G -> ... • 10G (or more) NICs are getting common on commodity servers • Transactions in 10G network • In the shortest packet case: • Maximum 14,880,952 packets/s*1 • Getting hard to handle in a single server... Ethernet Bandwidth and Transactions *1 shortest ethernet frame size 64bytes + preamble+IFG 20bytes = 84 bytes = 672 bits 10,000,000,000 / 672 = 14,880,952
- 7. 7Copyright © 2016 NTT Corp. All Rights Reserved. • UDP payload sizes • DNS • A/AAAA query: 40~ bytes • A/AAAA response: 100~ bytes • RADIUS • Access-Request: 70~ bytes • Access-Accept: 30~ bytes • Typically 100~ bytes with some attributes • In many cases 100~ bytes • 100 bytes transactions in 10G network • Max 7,530,120 transactions/s*1 • Less than shortest packet case, but still challenging How many transactions to handle? *1 100 bytes + IP/UDP/Ether headers 46bytes + preamble+IFG 20bytes = 166 bytes = 1328 bits 10,000,000,000 / 1328 = 7,530,120
- 8. 8Copyright © 2016 NTT Corp. All Rights Reserved. Basic technologies for network performance (not only for UDP)
- 9. 9Copyright © 2016 NTT Corp. All Rights Reserved. • TSO/GSO/GRO • Packet segmentation/aggregation • Reduce packets to process within server • Applicable to TCP*1 (byte stream) • Not applicable to UDP*2 (datagram) • UDP has explicit boundary between datagrams • Cannot segment/aggregate packets Basic technologies for network performance TCP UDP byte stream datagram TSO/GSO (segmentation) GRO (aggregation) MTU size Tx Server Rx Server :< *1 TCP in UDP tunneling (e.g. VXLAN) is OK as well *2 Other than UFO, which is rarely implemented on physical NICs Great performance gain! Not applicable
- 10. 10Copyright © 2016 NTT Corp. All Rights Reserved. • RSS • Scale network Rx processing in multi-core server • RSS itself is a NIC feature • Distribute packets to multi-queue in a NIC • Each queue has a different interrupt vector (Packets on each queue can be processed by different core) • Applicable to TCP/UDP • Common 10G NICs have RSS Basic technologies for network performance NIC rx queue rx queue Core Core packets RSS :) interrupt
- 11. 11Copyright © 2016 NTT Corp. All Rights Reserved. • 100 bytes UDP transaction performance • Measured by simple*1 (multi-threaded) echo server • OS: kernel 4.6.3 (in RHEL 7.2 environment) • Mid-range commodity server with 20 cores and 10G NIC: • NIC: Intel 82599ES (has RSS, max 64 queues) • CPU: Xeon E5-2650 v3 (2.3 GHz 10 cores) * 2 sockets Hyper-threading off (make analysis easy, enabled later) • Results: 270,000 transactions/s (tps) (approx. 360Mbps) • 3.6% utilization of 10G bandwidth Performance with RSS enabled NIC echo server thread thread thread x 20 UDP socket bulk 100bytes UDP packets*1 echo back *1 create as many threads as core num each thread just calls recvfrom() and sendto() *2 There is only 1 client (IP address). To spread UDP traffic on the NIC, RSS is configured to see UDP port numbers. This setting is not needed for common UDP servers.
- 12. 12Copyright © 2016 NTT Corp. All Rights Reserved. How to improve this?
- 13. 13Copyright © 2016 NTT Corp. All Rights Reserved. • sar -u ALL -P ALL 1 • softirq (interrupt processing) is performed only on NUMA Node 0, why? • although we have enough (64) queues for 20 cores... Identify bottleneck 19:57:54 CPU %usr %nice %sys %iowait %steal %irq %soft %guest %gnice %idle 19:57:54 all 0.37 0.00 42.58 0.00 0.00 0.00 50.00 0.00 0.00 7.05 19:57:54 0 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 19:57:54 1 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 19:57:54 2 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 19:57:54 3 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 19:57:54 4 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 19:57:54 5 1.82 0.00 83.64 0.00 0.00 0.00 0.00 0.00 0.00 14.55 19:57:54 6 0.00 0.00 87.04 0.00 0.00 0.00 0.00 0.00 0.00 12.96 19:57:54 7 0.00 0.00 85.19 0.00 0.00 0.00 0.00 0.00 0.00 14.81 19:57:54 8 0.00 0.00 85.45 0.00 0.00 0.00 0.00 0.00 0.00 14.55 19:57:54 9 0.00 0.00 85.19 0.00 0.00 0.00 0.00 0.00 0.00 14.81 19:57:54 10 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 19:57:54 11 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 19:57:54 12 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 19:57:54 13 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 19:57:54 14 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 19:57:54 15 1.82 0.00 83.64 0.00 0.00 0.00 0.00 0.00 0.00 14.55 19:57:54 16 0.00 0.00 87.04 0.00 0.00 0.00 0.00 0.00 0.00 12.96 19:57:54 17 1.82 0.00 83.64 0.00 0.00 0.00 0.00 0.00 0.00 14.55 19:57:54 18 0.00 0.00 85.45 0.00 0.00 0.00 0.00 0.00 0.00 14.55 19:57:54 19 0.00 0.00 85.45 0.00 0.00 0.00 0.00 0.00 0.00 14.55 Node 0 Node 1
- 14. 14Copyright © 2016 NTT Corp. All Rights Reserved. • RSS distributes packets to rx-queues • Interrupt destination of each queue is determined by /proc/irq/<irq>/smp_affinity • smp_affinity is usually set by irqbalance daemon softirq (interrupt processing) with RSS NIC rx queue rx queue Core Core packets RSS smp_affinity interrupt
- 15. 15Copyright © 2016 NTT Corp. All Rights Reserved. • smp_affinity*1 • irqbalance is using only Node 0 (cores 0-4, 10-14) • Can we change this? Check smp_affinity $ for ((irq=105; irq<=124; irq++)); do > cat /proc/irq/$irq/smp_affinity > done 01000 -> 12 -> Node 0 00800 -> 11 -> Node 0 00400 -> 10 -> Node 0 00400 -> 10 -> Node 0 01000 -> 12 -> Node 0 04000 -> 14 -> Node 0 00400 -> 10 -> Node 0 00010 -> 4 -> Node 0 00004 -> 2 -> Node 0 02000 -> 13 -> Node 0 04000 -> 14 -> Node 0 00001 -> 0 -> Node 0 02000 -> 13 -> Node 0 01000 -> 12 -> Node 0 00008 -> 3 -> Node 0 00800 -> 11 -> Node 0 00800 -> 11 -> Node 0 04000 -> 14 -> Node 0 00800 -> 11 -> Node 0 02000 -> 13 -> Node 0 *1 irq number can be obtained from /proc/interrupts
- 16. 16Copyright © 2016 NTT Corp. All Rights Reserved. • Some NIC drivers provide affinity_hint • affinity_hint is evenly distributed • To honor the hint, add "-h exact" option to irqbalance (via /etc/sysconfig/irqbalance, etc.)*1 Check affinity_hint $ for ((irq=105; irq<=124; irq++)); do > cat /proc/irq/$irq/affinity_hint > done 00001 -> 0 00002 -> 1 00004 -> 2 00008 -> 3 00010 -> 4 00020 -> 5 00040 -> 6 00080 -> 7 00100 -> 8 00200 -> 9 00400 -> 10 00800 -> 11 01000 -> 12 02000 -> 13 04000 -> 14 08000 -> 15 10000 -> 16 20000 -> 17 40000 -> 18 80000 -> 19 *1 If your NIC doesn't provide hint, you can use "-i" option or stop irqbalance to set their affinity manually
- 17. 17Copyright © 2016 NTT Corp. All Rights Reserved. • Added "-h exact" and restarted irqbalance • With hint honored, irqs are distributed to all cores Change irqbalance option $ for ((irq=105; irq<=124; irq++)); do > cat /proc/irq/$irq/smp_affinity > done 00001 -> 0 00002 -> 1 00004 -> 2 00008 -> 3 00010 -> 4 00020 -> 5 00040 -> 6 00080 -> 7 00100 -> 8 00200 -> 9 00400 -> 10 00800 -> 11 01000 -> 12 02000 -> 13 04000 -> 14 08000 -> 15 10000 -> 16 20000 -> 17 40000 -> 18 80000 -> 19
- 18. 18Copyright © 2016 NTT Corp. All Rights Reserved. • sar -u ALL -P ALL 1 • Though irqs looks distributed evenly, core 16-19 are not used for softirq... • Nodes look irrelevant this time Change irqbalance option 20:06:07 CPU %usr %nice %sys %iowait %steal %irq %soft %guest %gnice %idle 20:06:07 all 0.00 0.00 19.18 0.00 0.00 0.00 80.82 0.00 0.00 0.00 20:06:07 0 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 1 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 2 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 3 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 4 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 5 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 6 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 7 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 8 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 9 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 10 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 11 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 12 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 13 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 14 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 15 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:06:07 16 0.00 0.00 100.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 20:06:07 17 0.00 0.00 100.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 20:06:07 18 0.00 0.00 100.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 20:06:07 19 0.00 0.00 93.33 0.00 0.00 0.00 6.67 0.00 0.00 0.00 Node 0 Node 1
- 19. 19Copyright © 2016 NTT Corp. All Rights Reserved. • ethtool -S*1 • Revealed RSS has not distributed packets to queues 16-19 Check rx-queue stats $ ethtool -S ens1f0 | grep 'rx_queue_.*_packets' rx_queue_0_packets: 198005155 rx_queue_1_packets: 153339750 rx_queue_2_packets: 162870095 rx_queue_3_packets: 172303801 rx_queue_4_packets: 153728776 rx_queue_5_packets: 158138563 rx_queue_6_packets: 164411653 rx_queue_7_packets: 165924489 rx_queue_8_packets: 176545406 rx_queue_9_packets: 165340188 rx_queue_10_packets: 150279834 rx_queue_11_packets: 150983782 rx_queue_12_packets: 157623687 rx_queue_13_packets: 150743910 rx_queue_14_packets: 158634344 rx_queue_15_packets: 158497890 rx_queue_16_packets: 4 rx_queue_17_packets: 3 rx_queue_18_packets: 0 rx_queue_19_packets: 8 *1 Output format depends on drivers
- 20. 20Copyright © 2016 NTT Corp. All Rights Reserved. • RSS has indirection table which determines to which queue it spreads packets • Can be shown by ethtool -x • Only rx-queue 0-15 are used, 16-19 not used RSS Indirection Table $ ethtool -x ens1f0 RX flow hash indirection table for ens1f0 with 20 RX ring(s): 0: 0 1 2 3 4 5 6 7 8: 8 9 10 11 12 13 14 15 16: 0 1 2 3 4 5 6 7 24: 8 9 10 11 12 13 14 15 32: 0 1 2 3 4 5 6 7 40: 8 9 10 11 12 13 14 15 48: 0 1 2 3 4 5 6 7 56: 8 9 10 11 12 13 14 15 64: 0 1 2 3 4 5 6 7 72: 8 9 10 11 12 13 14 15 80: 0 1 2 3 4 5 6 7 88: 8 9 10 11 12 13 14 15 96: 0 1 2 3 4 5 6 7 104: 8 9 10 11 12 13 14 15 112: 0 1 2 3 4 5 6 7 120: 8 9 10 11 12 13 14 15 flow hash (hash value from packet header) rx-queue number
- 21. 21Copyright © 2016 NTT Corp. All Rights Reserved. • Change to use all 0-19? • This NIC's max rx-queues in the indirection table is actually 16 so we cannot use 20 queues • although we have 64 rx-queues... • Use RPS instead • Software emulation of RSS RSS Indirection Table # ethtool -X ens1f0 equal 20 Cannot set RX flow hash configuration: Invalid argument NIC rx queue rx queue Core Core packets RSS Core softirq backlog softirq backlog softirq backlog RPS softirq interrupt
- 22. 22Copyright © 2016 NTT Corp. All Rights Reserved. • This time I spread flows from rx-queue 6-9 to core 6-9 and 16-19 • Because they are all in Node 1 • rx-queue 6 -> core 6, 16 • rx-queue 7 -> core 7, 17 • rx-queue 8 -> core 8, 18 • rx-queue 9 -> core 9, 19 Use RPS # echo 10040 > /sys/class/net/ens1f0/queues/rx-6/rps_cpus # echo 20080 > /sys/class/net/ens1f0/queues/rx-7/rps_cpus # echo 40100 > /sys/class/net/ens1f0/queues/rx-8/rps_cpus # echo 80200 > /sys/class/net/ens1f0/queues/rx-9/rps_cpus
- 23. 23Copyright © 2016 NTT Corp. All Rights Reserved. • sar -u ALL -P ALL 1 • softirq is almost evenly distributed Use RPS 20:18:53 CPU %usr %nice %sys %iowait %steal %irq %soft %guest %gnice %idle 20:18:54 all 0.00 0.00 2.38 0.00 0.00 0.00 97.62 0.00 0.00 0.00 20:18:54 0 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 1 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 2 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 3 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 4 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 5 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 6 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 7 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 8 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 9 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 10 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 11 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 12 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 13 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 14 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 15 0.00 0.00 0.00 0.00 0.00 0.00 100.00 0.00 0.00 0.00 20:18:54 16 0.00 0.00 15.56 0.00 0.00 0.00 84.44 0.00 0.00 0.00 20:18:54 17 0.00 0.00 6.98 0.00 0.00 0.00 93.02 0.00 0.00 0.00 20:18:54 18 0.00 0.00 18.18 0.00 0.00 0.00 81.82 0.00 0.00 0.00 20:18:54 19 2.27 0.00 6.82 0.00 0.00 0.00 90.91 0.00 0.00 0.00
- 24. 24Copyright © 2016 NTT Corp. All Rights Reserved. • Now thanks to affinity_hint and RPS, we succeeded to spread flows almost evenly • Performance change • Before: 270,000 tps (approx. 360Mbps) • After: 17,000 tps (approx. 23Mbps) • Got worse... • Probably the reason is too heavy softirq • softirq is almost 100% in total • Need finer-grained profiling than sar RSS & affinity_hint & RPS
- 25. 25Copyright © 2016 NTT Corp. All Rights Reserved. • perf • Profiling tool developed in kernel tree • Identify hot spots by sampling CPU cycles • Example usage of perf • perf record -a -g -- sleep 5 • Save sampling results for 5 seconds to perf.data file • FlameGraph • Visualize perf.data in svg format • https://github.com/brendangregg/FlameGraph Profile softirq
- 26. 26Copyright © 2016 NTT Corp. All Rights Reserved. • FlameGraph of CPU0*1 • x-axis (width): CPU consumption • y-axis (height): Depth of call stack • queued_spin_lock_slowpath: lock is contended • udp_queue_rcv_skb: aquires socket lock Profile softirq *1 I filtered call stack under irq context from output of perf script to make the chart easier to see irq context is shown as "interrupt" here queued_spin_lock_slowpath udp_queue_rcv_skb
- 27. 27Copyright © 2016 NTT Corp. All Rights Reserved. • Echo server has only one socket bound to a certain port • softirq of each core pushes packets into socket queue concurrently • socket lock gets contended Socket lock contention NIC rx queue rx queue Core Core packets Core softirq backlog softirq backlog softirq backlog softirq UDP socket lock contention!! interrupt RPS
- 28. 28Copyright © 2016 NTT Corp. All Rights Reserved. • Split sockets by SO_REUSEPORT • Introduced by kernel 3.9 • SO_REUSEPORT allows multiple UDP sockets to bind the same port • One of the sockets is chosen on queueing each packet Avoid lock contention NIC rx queue rx queue Core Core packets Core softirq backlog softirq backlog softirq backlog UDP socket UDP socket UDP socket int on = 1; int sock = socket(AF_INET, SOCK_DGRAM, 0); setsockopt(sock, SOL_SOCKET, SO_REUSEPORT, &on, sizeof(on)); bind(sock, ...); socket selector *1 *1 select a socket by flow (packet header) hash by default interrupt RPS
- 29. 29Copyright © 2016 NTT Corp. All Rights Reserved. • sar -u ALL -P ALL 1 • CPU consumption in softirq became some more reasonable Use SO_REUSEPORT 20:44:33 CPU %usr %nice %sys %iowait %steal %irq %soft %guest %gnice %idle 20:44:34 all 3.26 0.00 37.23 0.00 0.00 0.00 59.52 0.00 0.00 0.00 20:44:34 0 3.33 0.00 28.33 0.00 0.00 0.00 68.33 0.00 0.00 0.00 20:44:34 1 3.33 0.00 25.00 0.00 0.00 0.00 71.67 0.00 0.00 0.00 20:44:34 2 1.67 0.00 23.33 0.00 0.00 0.00 75.00 0.00 0.00 0.00 20:44:34 3 3.28 0.00 32.79 0.00 0.00 0.00 63.93 0.00 0.00 0.00 20:44:34 4 3.33 0.00 33.33 0.00 0.00 0.00 63.33 0.00 0.00 0.00 20:44:34 5 1.69 0.00 23.73 0.00 0.00 0.00 74.58 0.00 0.00 0.00 20:44:34 6 3.28 0.00 50.82 0.00 0.00 0.00 45.90 0.00 0.00 0.00 20:44:34 7 3.45 0.00 50.00 0.00 0.00 0.00 46.55 0.00 0.00 0.00 20:44:34 8 1.69 0.00 37.29 0.00 0.00 0.00 61.02 0.00 0.00 0.00 20:44:34 9 1.67 0.00 33.33 0.00 0.00 0.00 65.00 0.00 0.00 0.00 20:44:34 10 1.69 0.00 18.64 0.00 0.00 0.00 79.66 0.00 0.00 0.00 20:44:34 11 3.23 0.00 35.48 0.00 0.00 0.00 61.29 0.00 0.00 0.00 20:44:34 12 1.69 0.00 27.12 0.00 0.00 0.00 71.19 0.00 0.00 0.00 20:44:34 13 1.67 0.00 21.67 0.00 0.00 0.00 76.67 0.00 0.00 0.00 20:44:34 14 1.67 0.00 21.67 0.00 0.00 0.00 76.67 0.00 0.00 0.00 20:44:34 15 3.33 0.00 35.00 0.00 0.00 0.00 61.67 0.00 0.00 0.00 20:44:34 16 6.67 0.00 68.33 0.00 0.00 0.00 25.00 0.00 0.00 0.00 20:44:34 17 5.00 0.00 65.00 0.00 0.00 0.00 30.00 0.00 0.00 0.00 20:44:34 18 6.78 0.00 54.24 0.00 0.00 0.00 38.98 0.00 0.00 0.00 20:44:34 19 4.92 0.00 63.93 0.00 0.00 0.00 31.15 0.00 0.00 0.00
- 30. 30Copyright © 2016 NTT Corp. All Rights Reserved. • before • after Userspace starts to work Use SO_REUSEPORT Interrupt processing (irq) Interrupt processing (irq) userspace thread (sys, user)
- 31. 31Copyright © 2016 NTT Corp. All Rights Reserved. • Perfomance change • RSS: 270,000 tps (approx. 360Mbps) • +affinity_hint+RPS: 17,000 tps (approx. 23Mbps) • +SO_REUSEPORT: 2,540,000 tps (approx. 3370Mbps) • Great improvement! • but... Use SO_REUSEPORT
- 32. 32Copyright © 2016 NTT Corp. All Rights Reserved. • More analysis • Socket lock is still contended Use SO_REUSEPORT queued_spin_lock_slowpath
- 33. 33Copyright © 2016 NTT Corp. All Rights Reserved. • SO_REUSEPORT uses flow hash to select queue by default • Same sockets can be selected by different cores • Socket lock still gets contended Socket lock contention again NIC rx queue rx queue Core1 Core2 packets Core0 softirq backlog softirq backlog softirq backlog UDP socket0 UDP socket1 UDP socket2 socket selector interrupt select queue by flow hash lock contention!! RPS
- 34. 34Copyright © 2016 NTT Corp. All Rights Reserved. • Select socket by core number • Realized by SO_ATTACH_REUSEPORT_CBPF/EBPF*1 • Introduced by kernel 4.5 • No lock contention between softirq • Usage • See example in kernel source tree • tools/testing/selftests/net/reuseport_bpf_cpu.c Avoid socket lock contention NIC rx queue rx queue Core1 Core2 packets Core0 softirq backlog softirq backlog softirq backlog UDP socket0 UDP socket1 UDP socket2 socket selector *1 BPF allows much more flexible logic but this time only cpu number is used interrupt select queue by core number RPS
- 35. 35Copyright © 2016 NTT Corp. All Rights Reserved. • before • after irq overhead gets less Use SO_ATTACH_REUSEPORT_EPBF Interrupt processing (irq) userspace thread (sys, user) Interrupt processing (irq) userspace thread (sys, user)
- 36. 36Copyright © 2016 NTT Corp. All Rights Reserved. • Perfomance change • RSS: 270,000 tps (approx. 360Mbps) • +affinity_hint+RPS: 17,000 tps (approx. 23Mbps) • +SO_REUSEPORT: 2,540,000 tps (approx. 3370Mbps) • +SO_ATTACH_...: 4,250,000 tps (approx. 5640Mbps) Use SO_ATTACH_REUSEPORT_EBPF
- 37. 37Copyright © 2016 NTT Corp. All Rights Reserved. • Userspace threads : sockets == 1 : 1 • No lock contention • But not necessarily on the same core as softirq • Pin userspace thread on the same core for better cache affinity • cgroup, taskset, pthread_setaffinity_np(), ... any way you like Pin userspace threads NIC rx queue rx queue Core1 Core2 packets Core0 softirq backlog softirq backlog softirq backlog UDP socket0 UDP socket1 UDP socket2 interrupt thread1 thread2 thread0 userspace kernel softirq NIC rx queue rx queue Core1 Core2 packets Core0 softirq backlog softirq backlog softirq backlog UDP socket0 UDP socket1 UDP socket2 interrupt thread0 thread1 thread2 userspace kernel softirq 1:1 pin RPS RPS
- 38. 38Copyright © 2016 NTT Corp. All Rights Reserved. • Perfomance change • RSS: 270,000 tps (approx. 360Mbps) • +affinity_hint+RPS: 17,000 tps (approx. 23Mbps) • +SO_REUSEPORT: 2,540,000 tps (approx. 3370Mbps) • +SO_ATTACH_...: 4,250,000 tps (approx. 5640Mbps) • +Pin threads: 5,050,000 tps (approx. 6710Mbps) Pin userspace threads
- 39. 39Copyright © 2016 NTT Corp. All Rights Reserved. • So far everything has been about Rx • No lock contention on Tx? Tx lock contention?
- 40. 40Copyright © 2016 NTT Corp. All Rights Reserved. • kernel has Qdisc (Queueing discipline) • Each Qdisc is linked to NIC tx-queue • Each Qdisc has its lock Tx queue Core Core Qdisc queue Qdisc queue thread thread threadCore NIC tx queue tx queue tx queueQdisc queue userspace kernel
- 41. 41Copyright © 2016 NTT Corp. All Rights Reserved. • By default Qdisc is selected by flow hash • Thus lock contention can happen • We haven't seen contention on Tx, why? Tx queue lock contention Core Core Qdisc queue Qdisc queue thread thread threadCore NIC tx queue tx queue tx queueQdisc queue userspace kernel select queue by flow hash lock contention!!
- 42. 42Copyright © 2016 NTT Corp. All Rights Reserved. • Because ixgbe (Intel 10GbE NIC driver) has an ability to set XPS automatically Avoid Tx queue lock contention $ for ((txq=0; txq<20; txq++)); do > cat /sys/class/net/ens1f0/queues/tx-$txq/xps_cpus > done 00001 -> core 0 00002 -> core 1 00004 -> core 2 00008 -> core 3 00010 -> core 4 00020 -> core 5 00040 -> core 6 00080 -> core 7 00100 -> core 8 00200 -> core 9 00400 -> core 10 00800 -> core 11 01000 -> core 12 02000 -> core 13 04000 -> core 14 08000 -> core 15 10000 -> core 16 20000 -> core 17 40000 -> core 18 80000 -> core 19
- 43. 43Copyright © 2016 NTT Corp. All Rights Reserved. • XPS allows kernel to select Tx queue (Qdisc) by core number • Tx has no lock contention XPS Core Core Qdisc queue Qdisc queue thread thread threadCore NIC tx queue tx queue tx queueQdisc queue userspace kernel select queue by core number
- 44. 44Copyright © 2016 NTT Corp. All Rights Reserved. • Try disabling it • Before: 5,050,000 tps (approx. 6710Mbps) • After: 1,086,000 tps (approx. 1440Mbps) How effective is XPS? # for ((txq=0; txq<20; txq++)); do > echo 0 > /sys/class/net/ens1f0/queues/tx-$txq/xps_cpus > done
- 45. 45Copyright © 2016 NTT Corp. All Rights Reserved. • XPS enabled • XPS disabled Disabling XPS Interrupt processing (irq) userspace thread Tx Interrupt processing (irq) userspace thread Tx userspace thread Rx userspace thread Rx queued_spin_lock_slowpath (lock contention)
- 46. 46Copyright © 2016 NTT Corp. All Rights Reserved. • Enable XPS again • Although ixgbe can automatically set XPS, not all drivers can do that • Make sure to check xps_cpus is configured Enable XPS # echo 00001 > /sys/class/net/<NIC>/queues/tx-0/xps_cpus # echo 00002 > /sys/class/net/<NIC>/queues/tx-1/xps_cpus # echo 00004 > /sys/class/net/<NIC>/queues/tx-2/xps_cpus # echo 00008 > /sys/class/net/<NIC>/queues/tx-3/xps_cpus ...
- 47. 47Copyright © 2016 NTT Corp. All Rights Reserved. • By making full use of multi-core with avoiding contention, we achieved • 5,050,000 tps (approx. 6710Mbps) • To get more performance, reduce overhead per core Optimization per core
- 48. 48Copyright © 2016 NTT Corp. All Rights Reserved. • GRO is enabled by default • Consuming 4.9% of CPU time Optimization per core dev_gro_receive
- 49. 49Copyright © 2016 NTT Corp. All Rights Reserved. • GRO is not applicable to UDP*1 • Disable it for UDP servers • WARNING: • Don't disable it if TCP performance matters • Disabling GRO makes TCP rx throughput miserably low • Don't disable it on KVM hypervisors as well • GRO boost throughput of tunneling protocol traffic as well as guest's TCP traffic on hypervisors GRO # ethtool -K <NIC> gro off *1 Other than UDP tunneling, like VXLAN
- 50. 50Copyright © 2016 NTT Corp. All Rights Reserved. • Perfomance change • RSS (+XPS): 270,000 tps (approx. 360Mbps) • +affinity_hint+RPS: 17,000 tps (approx. 23Mbps) • +SO_REUSEPORT: 2,540,000 tps (approx. 3370Mbps) • +SO_ATTACH_...: 4,250,000 tps (approx. 5640Mbps) • +Pin threads: 5,050,000 tps (approx. 6710Mbps) • +Disable GRO: 5,180,000 tps (approx. 6880Mbps) Disable GRO
- 51. 51Copyright © 2016 NTT Corp. All Rights Reserved. • iptables-related processing (nf_iterate) is performed • Although I have not added any rule to iptables • Consuming 3.00% of CPU time Optimization per core nf_iterate nf_iterate
- 52. 52Copyright © 2016 NTT Corp. All Rights Reserved. • With iptables kernel module loaded, even if you don't have any rules, it can incur some overhead • Some distributions load iptables module even when you don't add any rule • If you are not using iptables, unload the module iptables (netfilter) # modprobe -r iptable_filter # modprobe -r ip_tables
- 53. 53Copyright © 2016 NTT Corp. All Rights Reserved. • Perfomance change • RSS (+XPS): 270,000 tps (approx. 360Mbps) • +affinity_hint+RPS: 17,000 tps (approx. 23Mbps) • +SO_REUSEPORT: 2,540,000 tps (approx. 3370Mbps) • +SO_ATTACH_...: 4,250,000 tps (approx. 5640Mbps) • +Pin threads: 5,050,000 tps (approx. 6710Mbps) • +Disable GRO: 5,180,000 tps (approx. 6880Mbps) • +Unload iptables: 5,380,000 tps (approx. 7140Mbps) Unload iptables
- 54. 54Copyright © 2016 NTT Corp. All Rights Reserved. • On Rx, FIB (routing table) lookup is done twice • Each is consuming 1.82%~ of CPU time Optimization per core fib_table_lookup fib_table_lookup
- 55. 55Copyright © 2016 NTT Corp. All Rights Reserved. • One of two times of table lookup is for validating source IP addresses • Reverse path filter • Local address check • If you really don't need source validation, you can skip it FIB lookup on Rx # sysctl -w net.ipv4.conf.all.rp_filter=0 # sysctl -w net.ipv4.conf.<NIC>.rp_filter=0 # sysctl -w net.ipv4.conf.all.accept_local=1
- 56. 56Copyright © 2016 NTT Corp. All Rights Reserved. • Perfomance change • RSS (+XPS): 270,000 tps (approx. 360Mbps) • +affinity_hint+RPS: 17,000 tps (approx. 23Mbps) • +SO_REUSEPORT: 2,540,000 tps (approx. 3370Mbps) • +SO_ATTACH_...: 4,250,000 tps (approx. 5640Mbps) • +Pin threads: 5,050,000 tps (approx. 6710Mbps) • +Disable GRO: 5,180,000 tps (approx. 6880Mbps) • +Unload iptables: 5,380,000 tps (approx. 7140Mbps) • +Disable validation: 5,490,000 tps (approx. 7290Mbps) Disable source validation
- 57. 57Copyright © 2016 NTT Corp. All Rights Reserved. • Audit is a bit heavy when heavily processing packets • Consuming 2.5% of CPU time Optimization per core audit related processing
- 58. 58Copyright © 2016 NTT Corp. All Rights Reserved. • If you don't need audit, disable it Audit # systemctl disable auditd # reboot
- 59. 59Copyright © 2016 NTT Corp. All Rights Reserved. • Perfomance change • RSS (+XPS): 270,000 tps (approx. 360Mbps) • +affinity_hint+RPS: 17,000 tps (approx. 23Mbps) • +SO_REUSEPORT: 2,540,000 tps (approx. 3370Mbps) • +SO_ATTACH_...: 4,250,000 tps (approx. 5640Mbps) • +Pin threads: 5,050,000 tps (approx. 6710Mbps) • +Disable GRO: 5,180,000 tps (approx. 6880Mbps) • +Unload iptables: 5,380,000 tps (approx. 7140Mbps) • +Disable validation: 5,490,000 tps (approx. 7290Mbps) • +Disable audit: 5,860,000 tps (approx. 7780Mbps) Disable audit
- 60. 60Copyright © 2016 NTT Corp. All Rights Reserved. • IP ID field calculation (__ip_select_ident) is heavy • Consuming 4.82% of CPU time Optimization per core __ip_select_ident
- 61. 61Copyright © 2016 NTT Corp. All Rights Reserved. • This is an environment-specific issue • This happens if many clients has the same IP address • Cache contention by atomic operations • It is very likely you don't see this amount of CPU consumption without using tunneling protocol • If you really see this problem... • You can skip it only if you never send over-mtu-sized packets • Though it is very strict IP ID field calculation int pmtu = IP_PMTUDISC_DO; setsockopt(sock, IPPROTO_IP, IP_MTU_DISCOVER, &pmtu, sizeof(pmtu));
- 62. 62Copyright © 2016 NTT Corp. All Rights Reserved. • Perfomance change • RSS (+XPS): 270,000 tps (approx. 360Mbps) • +affinity_hint+RPS: 17,000 tps (approx. 23Mbps) • +SO_REUSEPORT: 2,540,000 tps (approx. 3370Mbps) • +SO_ATTACH_...: 4,250,000 tps (approx. 5640Mbps) • +Pin threads: 5,050,000 tps (approx. 6710Mbps) • +Disable GRO: 5,180,000 tps (approx. 6880Mbps) • +Unload iptables: 5,380,000 tps (approx. 7140Mbps) • +Disable validation: 5,490,000 tps (approx. 7290Mbps) • +Disable audit: 5,860,000 tps (approx. 7780Mbps) • +Skip ID calculation: 6,010,000 tps (approx. 7980Mbps) Skip IP ID calculation
- 63. 63Copyright © 2016 NTT Corp. All Rights Reserved. • So far we have not enabled hyper threading • It makes the number of logical cores 40 • Number of physical cores are 20 in this box • With 40 cores we need to rely more on RPS • Remind: Max usable rx-queues == 16 • Enable hyper-threading and set RPS on all rx- queues • queue 0 -> core 0, 20 • queue 1 -> core 1, 21 • ... • queue 10 -> core 10, 16, 30 • queue 11 -> core 11, 17, 31 • ... Hyper threading
- 64. 64Copyright © 2016 NTT Corp. All Rights Reserved. • Perfomance change • RSS (+XPS): 270,000 tps (approx. 360Mbps) • +affinity_hint+RPS: 17,000 tps (approx. 23Mbps) • +SO_REUSEPORT: 2,540,000 tps (approx. 3370Mbps) • +SO_ATTACH_...: 4,250,000 tps (approx. 5640Mbps) • +Pin threads: 5,050,000 tps (approx. 6710Mbps) • +Disable GRO: 5,180,000 tps (approx. 6880Mbps) • +Unload iptables: 5,380,000 tps (approx. 7140Mbps) • +Disable validation: 5,490,000 tps (approx. 7290Mbps) • +Disable audit: 5,860,000 tps (approx. 7780Mbps) • +Skip ID calculation: 6,010,000 tps (approx. 7980Mbps) • +Hyper threading: 7,010,000 tps (approx. 9310Mbps) • I guess more rx-queues would realize even better performance number Hyper threading
- 65. 65Copyright © 2016 NTT Corp. All Rights Reserved. • Tx Qdisc lock (_raw_spin_lock) is heavy • Not contended but involves many atomic operations • Being optimized in Linux netdev community More hot spots _raw_spin_lock
- 66. 66Copyright © 2016 NTT Corp. All Rights Reserved. • Memory alloc/free (slab) • Being optimized in netdev community as well More hot spots kmem/kmalloc/kfree
- 67. 67Copyright © 2016 NTT Corp. All Rights Reserved. • Virtualization • UDP servers as guests • Hypervisor can saturate CPUs or drop packets • We are going to investigate ways to boost performance in virtualized environment as well Other challenges
- 68. 68Copyright © 2016 NTT Corp. All Rights Reserved. • For 100bytes, we can achieve almost 10G • From: 270,000 tps (approx. 360Mbps) • To: 7,010,000 tps (approx. 9310Mbps) • Of course we need to take into account additional userspace work in real applications so this number is not applicable as is • To boost UDP performance • Applications (Most important!) • implement SO_REUSEPORT • implement SO_ATTACH_REUSEPORT_EBPF/CBPF • These are useful for TCP listening sockets as well • OS settings • Check smp_affinity • Use RPS if rx-queues are not enough • Make sure XPS is configured • Consider other tunings to reduce per-core overhead • Disable GRO • Unload iptables • Disable source IP validation • Disable auditd • Hardware • Use NICs which have enough RSS rx-queues if possible (as many queues as core num) Summary
- 69. 69Copyright © 2016 NTT Corp. All Rights Reserved. Thank you!