moongen
play

MoonGen A Scriptable High-Speed Packet Generator Paul Emmerich - PowerPoint PPT Presentation

Chair for Network Architectures and Services Technical University of Munich (TUM) MoonGen A Scriptable High-Speed Packet Generator Paul Emmerich January 31st, 2016 FOSDEM 2016 Chair for Network Architectures and Services Department of


  1. Chair for Network Architectures and Services Technical University of Munich (TUM) MoonGen A Scriptable High-Speed Packet Generator Paul Emmerich January 31st, 2016 FOSDEM 2016 Chair for Network Architectures and Services Department of Informatics Technical University of Munich (TUM) Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 1

  2. Chair for Network Architectures and Services Technical University of Munich (TUM) Outline Hardware vs. Software Packet Generators Architecture of MoonGen Hardware Timestamping on Commodity NICs Precise Rate Control Example Measurements Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 2

  3. Chair for Network Architectures and Services Technical University of Munich (TUM) Source: www.spirent.com Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 3

  4. Chair for Network Architectures and Services Technical University of Munich (TUM) Challenges for software packet generators ◮ Hardware packet generators are ◮ Precise ◮ Accurate ◮ Fast Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 4

  5. Chair for Network Architectures and Services Technical University of Munich (TUM) Challenges for software packet generators ◮ Hardware packet generators are ◮ Precise ◮ Accurate ◮ Fast ◮ Software packet generators ◮ Run on cheap commodity hardware ◮ Flexible Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 4

  6. Chair for Network Architectures and Services Technical University of Munich (TUM) Challenges for software packet generators ◮ Hardware packet generators are ◮ Precise ◮ Accurate ◮ Fast ◮ Software packet generators ◮ Run on cheap commodity hardware ◮ Flexible ◮ Key challenges for software packet generators ◮ Rate control ◮ Timestamping Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 4

  7. Chair for Network Architectures and Services Technical University of Munich (TUM) Design goals Design goal of MoonGen Combine the advantages of both approaches while avoiding their disadvantages. Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 5

  8. Chair for Network Architectures and Services Technical University of Munich (TUM) Design goals Design goal of MoonGen Combine the advantages of both approaches while avoiding their disadvantages. ◮ Fast: DPDK for packet I/O, explicit multi-core support ◮ Flexible: Craft all packets in user-controlled Lua scripts ◮ Timestamping: Utilize hardware features found on modern commodity NICs ◮ Rate control: Hardware features and a novel software approach Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 5

  9. Chair for Network Architectures and Services Technical University of Munich (TUM) Lua VM Architecture Lua VM Lua VM Userscript Userscript Userscript Userscript Userscript master spawn slave config API data API MoonGen MoonGen Core config API data API DPDK Q 0 ... Q n HW NIC NIC Port Port Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 6

  10. Chair for Network Architectures and Services Technical University of Munich (TUM) Hardware timestamping ◮ NICs support PTP for precise clock synchronization ◮ PTP support requires hardware timestamping capabilities ◮ These can be (mis-)used for delay measurements ◮ Typical precision ◮ ± 6.4 ns (Intel 10 GbE chips) ◮ ± 32 ns (Intel GbE chips) ◮ Some restrictions ◮ Packets must be UDP or PTP L2 protocol ◮ Minimum UDP packet size is 84 bytes Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 7

  11. Chair for Network Architectures and Services Technical University of Munich (TUM) Software rate control in existing packet generators DuT Loadgen p 5 NIC NIC p 5 p 4 p 3 p 2 p 1 p 0 Q memory Q NIC Wire ◮ Software tries to push single packets to the NIC ◮ Queues cannot be used, no batch processing ◮ NICs work with an asynchronous push-pull model ◮ This can lead to micro-bursts � Unreliable, imprecise, and bad performance Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 8

  12. Chair for Network Architectures and Services Technical University of Munich (TUM) Hardware rate control DuT HW rate control Loadgen enabled p 9 NIC NIC p 9 p 8 p 7 p 6 p 5 p 4 p 3 p 2 p 1 p 0 Q memory Q NIC Wire ◮ Modern NICs support rate control in hardware ◮ Limited to constant bit rate and bursty traffic ◮ Precision controlled by the hardware � High performance as queues can be used, but inflexible Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 9

  13. Chair for Network Architectures and Services Technical University of Munich (TUM) Software rate control based on invalid packets DuT HW rate control Loadgen disabled p 6 NIC NIC p i p i p i p i p i p 6 p 5 p 5 p 4 p 3 p 2 p 1 p 0 4 3 2 1 0 Q memory Q NIC Wire ◮ Fill gaps with invalid packets p i (e.g. bad CRC) ◮ NIC in the DuT drops invalid packets without side-effects ◮ Combines advantages of both approaches ◮ Precision limited by byte rate (0.8 ns per byte) and minimum packet size (33 byte) � High performance & high precision Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 10

  14. Chair for Network Architectures and Services Technical University of Munich (TUM) Does it work? ◮ Test setup: forward packets with Open vSwitch ◮ Measure the latency of the device under test 160 CBR Hardware (Median) 140 CBR Hardware (25/75th perc.) 120 Latency [ µ s] 100 80 60 40 20 0 0 0.5 1 1.5 2 Offered load [Mpps] Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 11

  15. Chair for Network Architectures and Services Technical University of Munich (TUM) Does it work? ◮ Compare both rate control approaches ◮ Maximum deviation: 2% 160 CBR Hardware (Median) 140 CBR Hardware (25/75th perc.) 120 CBR MoonGen (Median) Latency [ µ s] CBR MoonGen(25/75th perc.) 100 80 60 40 20 0 0 0.5 1 1.5 2 Offered load [Mpps] Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 12

  16. Chair for Network Architectures and Services Technical University of Munich (TUM) Does it matter? ◮ Compare CBR with Poisson traffic ◮ Different response from the device under test 160 CBR MoonGen (Median) 140 CBR MoonGen (25/75th perc.) 120 Poisson MoonGen (Median) Latency [ µ s] Poisson MoonGen (25/75th perc.) 100 80 60 40 20 0 0 0.5 1 1.5 2 Offered load [Mpps] Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 13

  17. Chair for Network Architectures and Services Technical University of Munich (TUM) Example: Linux NAPI ◮ Open vSwitch on Linux ◮ Uniform distribution caused by interrupt throttling 0.8 Probability [%] 0.6 0.4 0.2 0 0 10 20 30 40 50 60 70 Latency [ µ s] Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 14

  18. Chair for Network Architectures and Services Technical University of Munich (TUM) Example: Linux Virtualization (VirtIO) ◮ Open vSwitch forwarding through a VM ◮ Long tail distribution, typical for VMs 0.3 Probability [%] 0.2 0.1 0 100 150 200 250 300 350 Latency [ µ s] Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 15

  19. Chair for Network Architectures and Services Technical University of Munich (TUM) Example: Hardware Switch ◮ AS5712-54X 10/40 GbE OpenFlow switch ◮ Bimodal distribution caused by more input than output ports ◮ Some packets are forwarded directly (cut-through switch) ◮ Some packets are blocked by another flow and buffered Relative Probability [%] 6 4 2 0 1 1.5 2 2.5 3 3.5 4 Latency [ µ s] Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 16

  20. Chair for Network Architectures and Services Technical University of Munich (TUM) Summary ◮ Speeds of 10 Gbit/s per CPU core (64 byte packets) ◮ Sub-microsecond precision and accuracy ◮ Execute user-defined script code for each packet ◮ Easy to use Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 17

  21. Chair for Network Architectures and Services Technical University of Munich (TUM) Q & A Try MoonGen yourself! https://github.com/emmericp/MoonGen Questions? Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 18

  22. Chair for Network Architectures and Services Technical University of Munich (TUM) [Backup slide] Performance I: Lua can be faster than C ◮ UDP packets from varying source IP addresses Packet rate [Mpps] 15 10 MoonGen 5 Pktgen-DPDK 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 CPU frequency [GHz] ◮ Pktgen-DPDK needs a complicated main loop that covers all possibilites ◮ MoonGen can use a tight inner loop Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 19

  23. Chair for Network Architectures and Services Technical University of Munich (TUM) [Backup slide] Performance II: heavy workload and multi-core scaling ◮ Generate random UDP packets on 2 10 Gbit NICs ◮ 8 calls to Lua’s standard math.random per packet ◮ CPUs artificially clocked down to 1.2 GHz Packet rate [Mpps] 30 20 10 0 1 2 3 4 5 6 7 8 Number of 1.2 GHz CPU cores Paul Emmerich – MoonGen: A Scriptable High-Speed Packet Generator 20

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend


More recommend