Performance Comparison of 40G NFV Environments

undefined
A Host-based Performance Comparison of 
40G NFV Environments Focusing on 
Packet Processing Architectures and Virtual Switches
R. Kawashima
, S. Muramatsu, H. Nakayama
T. Hayashi, and H. Matsuo
Contents
Backgrounds
 NFV
 Related work
Packet Processing Architectures
Virtual Switches
Evaluation
Conclusion
1
Network Functions Virtualization (NFV)
2
Traditional
 
NFV
 
High cost
Low flexibility
Closed
Host Architectures in NFV
3
Packet Processing
Architecture
Packet Processing
Architecture
VM
PM
NAPI
DPDK
Netmap
Path-through
Bridge
OVS
VALE
Lagopus
vhost-net
vhost-user
NAPI
DPDK
Netmap
 
Which architecture or vswitch
should be used for NFV hosts ?
 
How much performance differs
depending on its architecture
and vswitch ?
VNF
Related Work
Performance of baremetal servers [1][2]
 They focused on performance bottlenecks
Performance of KVM and Container-based virtualization [3]
4
[1] P. Emmerich et al., “Assessing Soft- and Hardware Bottlenecks in PC-based Packet Forwarding Systems”, 
     Proc. ICN, pp. 78-83, 2015.
[2] S. Gallenmüller et al., “Comparison of Frameworks for High-Performance Packet IO”,
     Proc. ANCS, pp. 29-38, 2015.
[3] R. Bonafiglia et al., “Assessing the Performance of Virtualization Technologies for NFV: 
     a Preliminary Benchmarking”, Proc. EWSDN, pp. 67-72, 2015.
Our Evaluation
5
Combination of PM and VM architectures and vswitches
Throughput and Latency/Jitter
Intel and Mellanox 40GbE NICs
SR-IOV
Contents
Backgrounds
 NFV
 Related work
Packet Processing Architectures
Virtual Switches
Evaluation
Conclusion
6
Packet Processing Architecture
A way to forward packets (NIC <=> Applications)
 Interruption or Polling
 Single core or Multi cores
 Kernel space or User space
 Packet buffer structure
 Packet buffer management
7
The architecture has a major effect on the performance !
Three Architectures
8
NAPI
 
DMA
Netmap
DPDK
App
 
DMA
Netmap API
 
User
 
Kernel
 
DMA
DPDK API
Interruption
or
Polling
Polling
Interruption
or
Polling
Contents
Backgrounds
 NFV
 Related work
Packet Processing Architectures
Virtual Switches
Evaluation
Conclusion
9
A virtual switch bridges the host and VMs
Virtual Switch
10
L2 switching
(VLAN)
(OpenFlow)
(Tunneling)
(QoS)
The virtual switch has an impact on performance too !
Six Virtual Switches
11
Contents
Backgrounds
 NFV
 Related work
Packet Processing Architectures
Virtual Switches
Evaluation
Conclusion
12
Goals
13
Clarify performance characteristics of existing systems
Propose appropriate NFV host environments
Find a proper direction for performance improvement
Experiment 1 (Baremetal)
14
Server 1
Server 2
(DUT)
pktgen-dpdk
(Throughput)
OSTA
(Latency)
40GbE
vSwitch
PM
UDP traffic
eth1
eth2
eth1
eth2
Throughput
15
Intel XL710
Mellanox
ConnectX-3 EN
Throughputs with short packet sizes are far from wire rate
Throughputs differ depending on the NIC type
Latency
16
Intel XL710
Mellanox
ConnectX-3 EN
L2FWD-DPDK and Lagopus show worse latency
Jitter values are less than 10 μs
Experiment 2 (VM)
17
Server 1
Server 2
(DUT)
pktgen-dpdk
(Throughput)
OSTA
(Latency)
40GbE
UDP traffic
eth1
eth2
vSwitch
vSwitch
VM
PM
eth1
eth2
veth1
veth2
Throughput
18
Intel XL710
Mellanox
ConnectX-3 EN
NAPI/
vhost-net
DPDK/
vhost-user
The virtualization overhead is fairly large
Latency
19
Intel XL710
Mellanox
ConnectX-3 EN
NAPI/
vhost-net
DPDK/
vhost-user
The virtualization amplifies jitters
SR-IOV
20
Throughput
Latency
SR-IOV shows the best performance !
SR-IOV lacks flexibility of flow handling
Adequate NFV Host Environment
21
Throughput
Latency/Jitter
Usability
Intel XL710
Mellanox
ConnectX-3
DPDK/
vhost-user
OVS
Linux Bridge
L2FWD-DPDK
OVS-DPDK
NAPI/
vhost-net
NIC
Architecture/
Virtualization
Virtual Switch
Contents
Backgrounds
 NFV
 Related work
Packet Processing Architectures
Virtual Switches
Evaluation
Conclusion
22
Conclusion
Summary
 We have evaluated NFV host environments with 40GbE
 A NIC device affects performance characteristics
 DPDK should be used for both the host and the guest
 We cannot reach the wire rate with short packet sizes
 Virtualization worsens both throughput and latency
 SR-IOV showed better throughput and latency
Future Work
 Further evaluations
 VALE/Netmap based virtualization
 VALE and Lagopus on the VM
 Bidirectional and lots of flows
23
Slide Note
Embed
Share

This study compares the performance of 40G NFV environments focusing on packet processing architectures and virtual switches. It explores host architectures, NFV related work, evaluation of combinations of PM and VM architectures with different vswitches, and the impact of packet processing architecture on performance. Various aspects such as throughput, latency, jitter, and the use of Intel and Mellanox 40GbE NICs with SR-IOV are discussed.

  • NFV
  • Performance Comparison
  • Packet Processing
  • Virtualization
  • Virtual Switches

Uploaded on Oct 09, 2024 | 0 Views


Download Presentation

Please find below an Image/Link to download the presentation.

The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.

You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.

The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.

E N D

Presentation Transcript


  1. A Host-based Performance Comparison of 40G NFV Environments Focusing on Packet Processing Architectures and Virtual Switches R. Kawashima, S. Muramatsu, H. Nakayama T. Hayashi, and H. Matsuo

  2. Contents Backgrounds NFV Related work Packet Processing Architectures Virtual Switches Evaluation Conclusion 1

  3. Network Functions Virtualization (NFV) Traditional NFV Router vRouter VM VM FW vFW VM vEPC EPC HW-based SW-based High cost Low flexibility Closed Low cost High flexibility Open 2

  4. Host Architectures in NFV VM VNF NAPI DPDK Netmap Packet Processing Architecture vhost-net vhost-user Which architecture or vswitch should be used for NFV hosts ? Bridge OVS VALE Lagopus PM How much performance differs depending on its architecture and vswitch ? vSW NAPI DPDK Netmap Path-through Packet Processing Architecture 3

  5. Related Work Performance of baremetal servers [1][2] They focused on performance bottlenecks Performance of KVM and Container-based virtualization [3] [1] P. Emmerich et al., Assessing Soft- and Hardware Bottlenecks in PC-based Packet Forwarding Systems , Proc. ICN, pp. 78-83, 2015. [2] S. Gallenm ller et al., Comparison of Frameworks for High-Performance Packet IO , Proc. ANCS, pp. 29-38, 2015. [3] R. Bonafiglia et al., Assessing the Performance of Virtualization Technologies for NFV: a Preliminary Benchmarking , Proc. EWSDN, pp. 67-72, 2015. 4

  6. Our Evaluation Combination of PM and VM architectures and vswitches Throughput and Latency/Jitter Intel and Mellanox 40GbE NICs SR-IOV 5

  7. Contents Backgrounds NFV Related work Packet Processing Architectures Virtual Switches Evaluation Conclusion 6

  8. Packet Processing Architecture A way to forward packets (NIC <=> Applications) Interruption or Polling Single core or Multi cores Kernel space or User space Packet buffer structure Packet buffer management The architecture has a major effect on the performance ! 7

  9. Three Architectures NAPI Netmap DPDK vSW App DPDK API User Netmap API Kernel DPDK Netmap Rx Tx vSW PMD VALE Network Stack Network Stack Rx Tx Tx Rx Driver Interruption or Polling Interruption or Polling DMA Polling DMA DMA NIC NIC NIC 8

  10. Contents Backgrounds NFV Related work Packet Processing Architectures Virtual Switches Evaluation Conclusion 9

  11. Virtual Switch A virtual switch bridges the host and VMs VM VM L2 switching (VLAN) (OpenFlow) (Tunneling) (QoS) vSW Packet Processing Architecture The virtual switch has an impact on performance too ! NIC NIC 10

  12. Six Virtual Switches Name Running Space Kernel Kernel Kernel User User User Architecture Virtual I/O Support TAP/vhost-net TAP/vhost-net (QEMU) - vhost-user (to be supported) Linux Bridge OVS VALE L2FWD-DPDK OVS-DPDK Lagopus NAPI NAPI Netmap DPDK DPDK DPDK 11

  13. Contents Backgrounds NFV Related work Packet Processing Architectures Virtual Switches Evaluation Conclusion 12

  14. Goals Clarify performance characteristics of existing systems Propose appropriate NFV host environments Find a proper direction for performance improvement 13

  15. Experiment 1 (Baremetal) PM pktgen-dpdk (Throughput) OSTA (Latency) vSwitch UDP traffic eth1 eth1 eth2 eth2 Server 2 (DUT) Server 1 40GbE 14

  16. Throughput Mellanox ConnectX-3 EN Intel XL710 Throughputs differ depending on the NIC type Throughputs with short packet sizes are far from wire rate 15

  17. Latency Mellanox ConnectX-3 EN Intel XL710 L2FWD-DPDK and Lagopus show worse latency Jitter values are less than 10 s 16

  18. VM Experiment 2 (VM) vSwitch veth1 veth2 PM pktgen-dpdk (Throughput) OSTA (Latency) vSwitch UDP traffic eth1 eth2 eth1 eth2 Server 2 (DUT) Server 1 40GbE 17

  19. Throughput Mellanox ConnectX-3 EN Intel XL710 NAPI/ vhost-net The virtualization overhead is fairly large DPDK/ vhost-user 18

  20. Latency Mellanox ConnectX-3 EN Intel XL710 NAPI/ vhost-net The virtualization amplifies jitters DPDK/ vhost-user 19

  21. SR-IOV Throughput Latency SR-IOV shows the best performance ! SR-IOV lacks flexibility of flow handling 20

  22. Adequate NFV Host Environment Architecture/ Virtualization Virtual Switch NIC Linux Bridge Throughput NAPI/ vhost-net Intel XL710 OVS Latency/Jitter Mellanox ConnectX-3 DPDK/ vhost-user L2FWD-DPDK Usability OVS-DPDK 21

  23. Contents Backgrounds NFV Related work Packet Processing Architectures Virtual Switches Evaluation Conclusion 22

  24. Conclusion Summary We have evaluated NFV host environments with 40GbE A NIC device affects performance characteristics DPDK should be used for both the host and the guest We cannot reach the wire rate with short packet sizes Virtualization worsens both throughput and latency SR-IOV showed better throughput and latency Future Work Further evaluations VALE/Netmap based virtualization VALE and Lagopus on the VM Bidirectional and lots of flows 23

Related


More Related Content

giItT1WQy@!-/#giItT1WQy@!-/#giItT1WQy@!-/#giItT1WQy@!-/#giItT1WQy@!-/#