EmuEdge is a scalable, Hi-Fi, highly-automated real-world network emulator based on Xen/Linux Netns/OvS, supporting edge computing prototyping and general network emulation with both container/vm and can be hybridly combined with existing physical infrastructures. Ultimately, EmuEdge is designed to emulate real-world experiments with high Degree of Realism (DoR) in lab settings instead of costly physical field deployments. An overview of EmuEdge architecture is shown below:
If you want to learn more technical details, the following paper is associated with this code repository:
Zeng, Yukun, Mengyuan Chao, and Radu Stoleru. "EmuEdge: A hybrid emulator for reproducible and realistic edge computing experiments." 2019 IEEE International Conference on Fog Computing (ICFC). IEEE, 2019. (Download)
Biblatex entry:
@inproceedings{zeng2019emuedge,
title={EmuEdge: A hybrid emulator for reproducible and realistic edge computing experiments},
author={Zeng, Yukun and Chao, Mengyuan and Stoleru, Radu},
booktitle={2019 IEEE International Conference on Fog Computing (ICFC)},
pages={153--164},
year={2019},
organization={IEEE}
}
The following parameters can be configured to emulate/reproduce a realistic experiment:
- Network Topology: EmuEdge provides a JSON API based on adjacency list to define a network topology, with major network components virtualized in EmuEdge, we can easily replay a real-world topology as is on EmuEdge.
- Link Quality: Real-world networks are filled with noises and interferences, especially when wireless networks are gaining popularities nowadays. Comparing to previous work, EmuEdge provides more functionalities in emulating unrealiable networks. For each virtual link in EmuEdge, basic asymmetric/bidirectional network quality can be defined in multi perspectives of network (e.g., delay, packet loss). Moreover, with EmuEdge, we can define correlations in common network metrics to emulate network consistency and even reproduce a real-world network trace as is.
- Emulation Parameters (node configurations): Besides virtual routers and switches, currently EmuEdge supports two types of network nodes, container (linux netns) node for network-bounded emulation with high scalability and Xen VM node for computation-bounded experiments. The computing resources of EmuEdge server can be limited or dedicated to each node properly to achieve high DoR in computation with low costs.
- Mobility/Synthetic trace (in progress): EmuEdge aims to support synthetic traces (from traditional simulators) and mobility emulation in future work. The figure below shows an example of extracting information from real experiment for repeated experiment replays on EmuEdge.
The most intuitive way of interacting with EmuEdge is to define an edge topology through JSON adjacency list. Starting such a customized topology is simple:
[root@xenserver ~]# cd /PATH/TO/hybridnet
[root@xenserver hybridnet]# python
Python 2.7.5 (default, Nov 20 2015, 02:00:19)
[GCC 4.8.5 20150623 (Red Hat 4.8.5-4)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import xen
>>> xnet=xen.test_topo(topo='exps/twoway_simple.topo', start=True, nolog=False)
The start
parameter controls if all devices will be booted automatically after setting up the topology, nolog
parameter gives us the ability to select between fast no-log mode or debugging mode that shows useful logs.
A twoway_simple
topology configuration with DHCP and traffic shaping for network full-duplexity test can be found in the Appendix part, for more useful examples with probably routing, NAT and other advanced capabilities, pls. check out topos/
.
Through the script ee
in our root directory, we can easily interact with containers in EmuEdge. Every container hosts can run real programs originally on your system as is like in MiniNet, an simple example as follows:
[root@xenserver hybridnet]# ./ee h1 ip addr
1: lo: <LOOPBACK> mtu 65536 qdisc noop state DOWN qlen 1
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
389: h1-in0@if388: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc tbf state UP qlen 1000
link/ether 0a:48:5b:51:35:dc brd ff:ff:ff:ff:ff:ff link-netnsid 0
inet 10.0.0.2/24 scope global h1-in0
valid_lft forever preferred_lft forever
inet6 fe80::848:5bff:fe51:35dc/64 scope link
valid_lft forever preferred_lft forever
To interact with VM hosts, we can either do it with a Xen management console, or ssh
to it through a container hosts that has connectivity with it. Currently, EmuEdge is limited in fully controlling VMs in terms of IP address, DHCP servers will assign random addresses which adds up the complexity of interacting with them. This can be probably resolved in the future with methods proposed in issue #13.
After initing the topology, we can still modify the topology with EmuEdge Python APIs. For example, shutting down a VM with id $node_id
:
xnet.shutdown_node($node_id)
Creating a new Android VM with Android
snapshot (if it is present in your XenServer), named android-test
, overriding the default configuration with 2 cpus and 2048MB memory:
xnet.create_new_dev("Android", "android-test", True, vcpu=2, mem=2048)
We left a more comprehensive API documents for interactive control over the topology as future work.
-
bean/: EmuEdge core objects (e.g., vm, container, interface, etc) and functions
-
utils/: helper/non-core utility functions
-
trace/: network trace analysis module based on iproute2 and a sample real-world wifi trace
-
bash/: some example scripts for manipulating virtual networks
-
docs/: paper manuscript and referred figures go here
-
topo/: testing topology sets covering all functions of EmuEdge, including paper experiments topologies
-
ee: container operation entry
-
xen.py: main entry for EmuEdge
- Xen and XenAPI
- Snapshots/Templates for the OSs to emulate in the target system
- Python 2
- Open vSwitch
- Linux iproute2
All our tests are conducted with Python 2.7.5 on XenServer 7.1.0, OvS version 2.3.2. If you are using also XenServer 7.1.0, all the other dependencies are by default installed.
- Dnsmasq-2.79, for DHCP configurations, especially helpful in setting up VM connectivities.
- Linux tc, tbf, netem, iptables, for traffic shaping, rate limiting and routing support.
[
{
"id":0,
"name":"xswitch0",
"type":0,
"neighbors":[
{
"id":1
},
{
"id":2,
"link_control":{
"rate":{
"latency":"1ms",
"burst":"47.5kb",
"rate":"95Mbit"
}
}
},
{
"id":3,
"link_control":{
"rate":{
"latency":"1ms",
"burst":"47.5kb",
"rate":"95Mbit"
}
}
},
{
"id":4,
"link_control":{
"rate":{
"latency":"1ms",
"burst":"47.5kb",
"rate":"95Mbit"
}
}
},
{
"id":5,
"link_control":{
"rate":{
"latency":"1ms",
"burst":"47.5kb",
"rate":"95Mbit"
}
}
}
]
},
{
"id":1,
"name":"prouter0",
"type":3,
"nat":{
"is_open":false
},
"dhcp":{
"is_open":true,
"if":0,
"range_low":"10.0.0.50",
"range_high":"10.0.0.254"
},
"ifs":[
{
"id":0,
"ip":"10.0.0.1/24"
}
],
"neighbors":[
{
"if":0,
"id":0
}
]
},
{
"id":2,
"name":"centos1",
"type":1,
"image":"tcentos",
"vcpus":2,
"mem":2048,
"override":false,
"neighbors":[
{
"id":0,
"link_control":{
"rate":{
"latency":"1ms",
"burst":"47.5kb",
"rate":"95Mbit"
}
}
}
]
},
{
"id":3,
"name":"centos2",
"type":1,
"image":"tcentos",
"vcpus":2,
"mem":2048,
"override":false,
"neighbors":[
{
"id":0,
"link_control":{
"rate":{
"latency":"1ms",
"burst":"47.5kb",
"rate":"95Mbit"
}
}
}
]
},
{
"id":4,
"name":"h1",
"type":3,
"nat":{
"is_open":false
},
"dhcp":{
"is_open":false
},
"ifs":[
{
"id":0,
"ip":"10.0.0.2/24"
}
],
"neighbors":[
{
"if":0,
"id":0,
"link_control":{
"rate":{
"latency":"1ms",
"burst":"47.5kb",
"rate":"95Mbit"
}
}
}
]
},
{
"id":5,
"name":"h2",
"type":3,
"nat":{
"is_open":false
},
"dhcp":{
"is_open":false
},
"ifs":[
{
"id":0,
"ip":"10.0.0.3/24"
}
],
"neighbors":[
{
"if":0,
"id":0,
"link_control":{
"rate":{
"latency":"1ms",
"burst":"47.5kb",
"rate":"95Mbit"
}
}
}
]
}
]