Session based high bandwidth throughput testing.
RP2System and Network Engineering System and Network Engineering
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Student: Bram ter Borch
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Research Question2
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
UDP vs TCP3
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
UDP
connectionless
8 byte header
no ordering
lightweight
TCP
connection oriented
20 byte header
ordering
heavyweight
reliable
congestion control
The Channel
Research Question4
What is needed to perform high bandwidth session based throughput tests and how to go
beyond pure network infrastructure testing? 010101010101101010101110110100010010110111101101010101001001010010100100010010010100101010100101001010100101010010101001010100101010100101010101010101001010100101001010101010101011100101011011000100101011010101101100011001101001001001110100101001010010101001001001
The term ”high bandwidth” references to at least 40Gb/s.
The term ”session based” references to TCP traffic.
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Test environment5
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
A
B
C
ED
S
40Gb/s
100Gb/s 40Gb/s100Gb/s
40Gb/s
MGMT M
100Gb/s
1Gb/s1Gb/s
1Gb/s1Gb/s
Vlan100:10.10.10.0/24
Vlan 66:10.60.66.0/24
Internet
4x10Gb/s
Tools6
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
BoNeSi
pktgen-kernelMoonGen
pktgen-DPDK
First important result7
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
DPDK is the way to go
Use cases8
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
NR Use case DUT Explanation
UC1 Bandwidth generation Client The goal is to see if the client is capable of filling up the link
and to reach the maximum amount of pps
UC2 Throughput Switch/RouterGenerate the maximum amount of bidirectional data to make sure the intermediate hardware is able to forward at line rate
UC3 Session per second Client/Server Get the hardware limitations of the end hosts.
UC4 Application Server and intermediate devices
The clients will try to overload the server with requests at application level
Experiment UC1 bandwidth9
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Experiment UC1 PPS10
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Experiment UC3 HTTP link usage11
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Experiment UC3 sessions12
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Real world test13
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
AE11
24x
10G
b/s
AE11
34x
10G
b/s
Router1BRouter1A
Firewall1 Firewall2
DC1-1A DC1-1B DC2-1A DC2-1B
SURFnetVCP
5x10Gb/s
100Gb/s
40Gb/s
40G
b/s
40Gb/s
40Gb/s
SERVER A
40Gb/s
40Gb/s
40Gb/s40Gb/s
2x40Gb/s
Active - passive
40Gb/s
2x40Gb/s
AE0 AE0
sync
100Gb/s
Router
4x10
Gb/
s
QFX10k2
CLIENT
40G
b/s
A P
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Real world result - bandwidth14
Student: Bram ter Borch
NGINX RAW TCP HTTP
Conclusion15
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
DPDK is the way to go for high bandwidth session based throughput testing
The use cases are valid
The weakest link in a path can be found
Monitoring is very important
Future work16
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
DPDK tests have to be run using 100Gb/s interfaces
IPv6 support has to be added to WARP
Other layer 7 protocols have to be added to WARP
Thank you
Questions
010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
Student: Bram ter Borch010101010101101010101110110100010010110111101101100101011011000100101011010101101100011001101001001001110100101001010010101001001001
