Performance Measurements of Linux, DanOS, VYOS, VPP, and Linux XDP at 100GE
Tests still being performed - check back at the end of March 2022 for final results
Results:
Table below represent Millions of Packets Per Second (MPPS) send for forwarding via the router software vs packetloss of the end destination of expected packets.
Note:
- droptest % represents the % of loss of the legitimate packets
- some higher rates not tested once significant loss was demonstrated at lower levels
- no optimisations performed on these routers unless otherwise noted below
Single-flow
MPPS | 1.5 | 3 | 4.5 | 6 | 7.5 | 9 | 10 | 12 | 15 | 18 | 20 | 30 |
Danos | 0.0% | 0.0% | 0.0% | 0.1% | 0.1% | 0.0% | 11.2% | 27.3% | 42.0% | 72.0% | ||
VPP | ||||||||||||
Linux | ||||||||||||
LinXDP | ||||||||||||
VYOS |
Single-flow w/ 5 blocking rules
MPPS | 1.5 | 3 | 4.5 | 6 | 7.5 | 9 | 10 | 12 | 15 | 18 | 20 | 30 |
Danos | 0.0% | 0.0% | 0.0% | 0.0% | 12.4% | |||||||
VPP | ||||||||||||
Linux | ||||||||||||
LinXDP | ||||||||||||
VYOS |
Single-flow w/ 50 blocking rules
MPPS | 1.5 | 3 | 4.5 | 6 | 7.5 | 9 | 10 | 12 | 15 | 18 | 20 | 30 |
Danos | 0.0% | 0.0% | 0.1% | 0.1% | 13.7% | 31.1% | 40.3% | |||||
VPP | ||||||||||||
Linux | ||||||||||||
LinXDP | ||||||||||||
VYOS |
Single-flow w/ 50 blocking rules, GRE tunnel forwarding outcome
MPPS | 1.5 | 3 | 4.5 | 6 | 7.5 | 9 | 10 | 12 | 15 | 18 | 20 | 30 |
Danos | 0.0% | 0.0% | 10.8% | 34.3% | 47.8% | 56.7% | 63.1% | |||||
VPP | ||||||||||||
Linux | ||||||||||||
LinXDP | ||||||||||||
VYOS |
Multiple-flow w/ 50 blocking rules
MPPS | 1.5 | 3 | 4.5 | 6 | 7.5 | 9 | 10 | 12 | 15 | 18 | 20 | 30 |
Danos | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 15.0% | 27.2% | |
VPP | ||||||||||||
Linux | ||||||||||||
LinXDP | ||||||||||||
VYOS |
Multiple-flow w/ 50 blocking rules, 900K routes
MPPS | 1.5 | 3 | 4.5 | 6 | 7.5 | 9 | 10 | 12 | 15 | 18 | 20 | 30 |
Danos | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.1% | 15.1% | 27.0% | 49.2% |
VPP | ||||||||||||
Linux | ||||||||||||
LinXDP | ||||||||||||
VYOS |
Multiple-flow w/ 50 blocking rules, 900K routes, 50% of traffic dropped
MPPS | 1.5 | 3 | 4.5 | 6 | 7.5 | 9 | 10 | 12 | 15 | 18 | 20 | 30 |
Danos | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 10.6% | 22.5% | 47.2% |
VPP | ||||||||||||
Linux | ||||||||||||
LinXDP | ||||||||||||
VYOS |
Test Environment & pktgen tool
Network Card:
mlx5_core 0000:3b:00.1: firmware version: 16.27.6120
mlx5_core 0000:3b:00.1: 126.016 Gb/s available PCIe bandwidth, limited by 8.0 GT/s PCIe x16 link at 0000:3a:00.0 (capable of 252.048 Gb/s with 16.0 GT/s PCIe x16 link)
We wont go into building pktgen as there's plenty of doco out there on this. Just for reference purposes on how we ran pktgen:
LD_LIBRARY_PATH=/usr/local/lib64/ /root/pktgen-dpdk/usr/local/bin/pktgen -l 2,4,6 -n 2 -a 3b:00.1 -d librte_net_mlx5.so -- -p 0x1 -P -m "[4:6].0"
Traffic generated:
Static destination MAC (The test Target)
pktgen:
set 0 rate 10 (This is % of 100GE, adjusted accordingly at 1% = 1.5MPPS)
set 0 size 64
set 0 count 50000000
set 0 proto udp
set 0 dst ip 10.22.23.102
set 0 src ip 10.22.22.101/24
set 0 dst mac XX:XX:XX:XX:d1:7b
set 0 src mac XX:XX:XX:XX:36:75
set 0 type ipv4
Single flow:
Single target IP address
UDP traffic, 64 bytes per packet, same src/dst ports
No firewall policies
Single flow, 5 firewall policies:
Single target IP address
UDP traffic, 64 bytes per packet, same src/dst ports
5 firewall policies
Single flow, 50 firewall policies:
Single target IP address
UDP traffic, 64 bytes per packet, same src/dst ports
50 firewall policies
Single flow, GRE tunnel:
Single target IP address
UDP traffic, 64 bytes per packet, same src/dst ports
50 firewall policies
GRE encap traffic and forward to static destination
Multiple flow:
254 destination IP addresses (multi-flow)
UDP traffic, 64 bytes per packet, random src ports (multi-flow)
50 firewall policies
pktgen:
range 0 src port 53 53 1000 1
range 0 dst ip 10.22.23.1 10.22.23.1 10.22.23.254 0.0.0.1
range 0 src ip 10.22.22.101 10.22.22.101 10.22.22.101 0.0.0.0
range 0 src mac XX:XX:XX:XX:36:75 XX:XX:XX:XX:36:75 XX:XX:XX:XX:36:75 00:00:00:00:00:00
range 0 dst mac XX:XX:XX:XX:d1:7b XX:XX:XX:XX:d1:7b XX:XX:XX:XX:d1:7b 00:00:00:00:00:00
enable 0 range
900K Route Test:
254 destination IP addresses
UDP traffic, 64 bytes per packet, random src/dst ports
50 firewall policies
900K loaded route table
Drop Test:
1 destination IP addresses, multiple ports (multi-flow)
UDP traffic, 64 bytes per packet, random src/dst ports
50 firewall policies, default deny
900K loaded route table
Half test traffic configured to be dropped
DanOS
Version: 2105
Built on: Fri Jun 11 11:58:32 UTC 2021
HW Model: PowerEdge R440
CPU: Intel Xeon Silver 4210R CPU @ 2.4Ghz
set protocols static arp 10.22.22.101 hwaddr 'XX:XX:XX:XX:XX:XX'
set protocols static arp 10.22.22.101 interface dp0p59s0f1
set protocols static route 10.22.23.0/24 next-hop 10.22.22.101
Firewall policies:
set security ip-packet-filter group ipv4 ip-version ipv4
set security ip-packet-filter group ipv4 rule 1 action drop
set security ip-packet-filter group ipv4 rule 1 match source ipv4 host 1.1.1.1
set security ip-packet-filter group ipv4 rule 2 action drop
set security ip-packet-filter group ipv4 rule 2 match source ipv4 host 1.1.2.1
set security ip-packet-filter group ipv4 rule 3 action drop
set security ip-packet-filter group ipv4 rule 3 match source ipv4 host 1.1.3.1
...etc...
set security ip-packet-filter interface dp0p59s0f1 in ipv4
For GRE test:
set interfaces tunnel tun0 address 10.90.4.102/24
set interfaces tunnel tun0 encapsulation gre
set interfaces tunnel tun0 local-ip 10.22.22.102
set interfaces tunnel tun0 remote-ip 10.22.22.101