1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
|
NFV Service Density
-------------------
Network Function Virtualization (NFV) service density tests focus on
measuring total per server throughput at varied NFV service “packing”
densities with vswitch providing host dataplane. The goal is to compare
and contrast performance of a shared vswitch for different network
topologies and virtualization technologies, and their impact on vswitch
performance and efficiency in a range of NFV service configurations.
Each NFV service instance consists of a set of Network Functions (NFs),
running in VMs (VNFs) or in Containers (CNFs), that are connected into a
virtual network topology using VPP vswitch running in Linux user-mode.
Multiple service instances share the vswitch that in turn provides per
service chain forwarding context(s). In order to provide a most complete
picture, each network topology and service configuration is tested in
different service density setups by varying two parameters:
- Number of service instances (e.g. 1,2,4..10).
- Number of NFs per service instance (e.g. 1,2,4..10).
The initial implementation of NFV service density tests in
|csit-release| is using two NF applications:
- VNF: DPDK L3fwd running in KVM VM, configured with /8 IPv4 prefix
routing. L3fwd got chosen as a lightweight fast IPv4 VNF application,
and follows CSIT approach of using DPDK sample applications in VMs for
performance testing.
- CNF: VPP running in Docker Container, configured with /24 IPv4 prefix
routing. VPP got chosen as a fast IPv4 NF application that supports
required memif interface (L3fwd does not). This is similar to all
other Container tests in CSIT that use VPP.
Tests are designed such that in all tested cases VPP vswitch is the most
stressed application, as for each flow vswitch is processing each packet
multiple times, whereas VNFs and CNFs process each packets only once. To
that end, all VNFs and CNFs are allocated enough resources to not become
a bottleneck.
Service Configurations
~~~~~~~~~~~~~~~~~~~~~~
Following NFV network topologies and configurations are tested:
- VNF Service Chains (VSC) with L2 vswitch
- *Network Topology*: Sets of VNFs dual-homed to VPP vswitch over
virtio-vhost links. Each set belongs to separate service instance.
- *Network Configuration*: VPP L2 bridge-domain contexts form logical
service chains of VNF sets and connect each chain to physical
interfaces.
- CNF Service Chains (CSC) with L2 vswitch
- *Network Topology*: Sets of CNFs dual-homed to VPP vswitch over
memif links. Each set belongs to separate service instance.
- *Network Configuration*: VPP L2 bridge-domain contexts form logical
service chains of CNF sets and connect each chain to physical
interfaces.
- CNF Service Pipelines (CSP) with L2 vswitch
- *Network Topology*: Sets of CNFs connected into pipelines over a
series of memif links, with edge CNFs single-homed to VPP vswitch
over memif links. Each set belongs to separate service instance.
- *Network Configuration*: VPP L2 bridge-domain contexts connect each
CNF pipeline to physical interfaces.
Thread-to-Core Mapping
~~~~~~~~~~~~~~~~~~~~~~
CSIT defines specific ratios for mapping software threads of vswitch and
VNFs/CNFs to physical cores, with separate ratios defined for main
control threads and data-plane threads.
In |csit-release| NFV service density tests run on Intel Xeon testbeds
with Intel Hyper-Threading enabled, so each physical core is associated
with a pair of sibling logical cores corresponding to the hyper-threads.
|csit-release| executes tests with the following software thread to
physical core mapping ratios:
- vSwitch
- Data-plane on single core
- (data:core) = (1:1) => 2dt1c - 2 Data-plane Threads on 1 Core.
- (main:core) = (1:1) => 1mt1c - 1 Main Thread on 1 Core.
- Data-plane on two cores
- (data:core) = (1:2) => 4dt2c - 4 Data-plane Threads on 2 Cores.
- (main:core) = (1:1) => 1mt1c - 1 Main Thread on 1 Core.
- VNF and CNF
- Data-plane on single core
- (data:core) = (1:1) => 2dt1c - 2 Data-plane Threads on 1 Core per
NF.
- (main:core) = (2:1) => 2mt1c - 2 Main Threads on 1 Core, 1 Thread
per NF, core shared between two NFs.
Maximum tested service densities are limited by a number of physical
cores per NUMA. |csit-release| allocates cores within NUMA0. Support for
multi NUMA tests is to be added in future release.
|
