Performance Analysis of VXLAN and NVGRE Tunneling Protocol on Virtual Network

Bulletin of Electrical Engineering and Informatics
ISSN: 2302-9285
Vol. 6, No. 3, September 2017, pp. 295~300, DOI: 10.11591/eei.v6i3.622  295
Received June 10, 2017; Revised July 13, 2017; Accepted August 27, 2017
Performance Analysis of VXLAN and NVGRE Tunneling
Protocol on Virtual Network
Mirdan Syahid Mulya Sudrajat*
1
, Doan Perdana
2
, Ridha Muldina Negara
3
School of Electrical Engineering, Telkom University, Telekomunikasi street no.1, Bandung,
West Java, Indonesia
*Corresponding author, e-mail: mirdsmulya@gmail.com
1
, doanperdana@telkomuniversity.ac.id
2
,
ridhanegara@telkomuniversity.ac.id
3
Abstract
Virtualization is a new revolutionary approach in networking industry, its make possible to build
several virtual machine (VM) in one physical hardware. In virtualization practice, one VM might be
connected to others, but not all of VM in one environment must be connected due the privacy and security
issues. One of the solutions which can address this issue is tunneling protocol. Tunneling protocol is a
layer-2-in-layer-3 protocol which can isolate tenant traffic in virtualize environment. This research
conducted about the performance of VXLAN and NVGRE tunneling protocol which works on virtualize
environment and aims to determine the perfomances of throughput, delay, jitter, and vCPU Usage using
variable packet size in range of 128-1514 byte. From the the result, can be conclude that both of tunneling
protocol can isolate the traffic between tenant. For the performance result, NVGRE has the highest value
of throughput, 771,02 Mbps and the VXLAN got 753,62 Mbps. For the delay NVGRE got 2.24 ms and
VXLAN got 2.29 ms. For the jitter, NVGRE has better rate value of 0.361 ms, than VXLAN value of 0.348
ms, and the vCPU usage performance, NVGRE has the highest performance too that value is 60.57%. So
on overall performance NVGRE has the better performance than VXLAN.
Keywords: tunneling protocol, network virtualization, overlay network
1. Introduction
In this era of global competition, the availability of cloud computing infrastructure in a
corporate or agency environment becomes very important. In practice, there are often occurs
problems faced. Those problems are the scalability and flexibility of the cloud computing
infrastructure [1]. As the massive ICT development, one the emerging solution to simplify these
problems is approaching virtualization technology. Which has been widely implemented by
small and large scale ICT companies like datacenter [1]. This virtualization technology can’t be
denied that can decreasing the CAPEX and OPEX side [2].
Virtualization is the use of software to allocate the resources of hardware into several
parts like virtual machine (VM). With virtualization, the network complexity can be simplified, and
can be deployed according to user needs flexibly and scalable [3]. In datacenter-scale
virtualization, which can accommodate a large number of tenants. Each tenant may be has
more than one integrated VM to execute an application, so each tenant must be isolated from
others [4]. One of the solution that arises to address those problems is a tunneling protocol
technology. This protocol is able to isolate traffic within the virtualization technology which called
overlay virtual network (OVN) [5]. There are also another way that use of VLAN technology
(802.1Q). However, this technology is considered that not reliable enough because of scalability
issue. VLAN relatively low if its compare with the tunneling protocol technology, because VLAN
only has the capacity of VLAN tag around 4000 VLAN ID.
Tunneling protocols that works in virtual environments or called with overlay virtual
network, has three protocols, namely VXLAN, NVGRE, and STT [6]. These tunneling protocols
can isolate tenant traffic up to 16 million tunnel ids [6], coupled with a multitenancy features that
increase tenant flexibility. But in implementation of this technology there are some shortcoming
issues, which can burden of vCPU usage and bandwidth effectiveness, these because the
addition of encapsulation process to the packet before its transmitted, so the vCPU usage
needs to work more than before, along with the increasing the packet size as well [6]. At this

 ISSN: 2302-9285
Bulletin of EEI Vol. 6, No. 3, September 2017 : 295 – 300
296
research will analyze the performance comparison of the two tunneling protocols, there are
NVGRE and VXLAN. It aims to see which one has the better performance.
2. Research Method
This research used system architecture as shown in Figure 1. and the topology shown
in Figure 2. which consists of two servers, user PC and one switch. Each server is built in a
virtual environment using the Xen Server hypervisor and thereafter built one VM from Ubuntu
Linux as host. User PC is used as hypervisor manager by installing Xen Center which is the
default application of Xen Hypervisor that serves to control and monitoring virtual resources
remotely. Once the system is ready, the tunneling protocol is built using bash scripting on each
VM that has previously been installed Open vSwitch. Open vSwitch is use as a tunnel end point
to implementing this tunneling protocol at the system [7].
Figure 1. Tunneling system architecture
There are four test scenarios performed to see the performance of the tunneling
protocol VXLAN and NVGRE, which work in virtual environment [8]. In addition, the two of
tunneling protocols were also tested native-based communication (non-tunneling), as a
comparison with the two tunneling protocols tested. There are 4 parameters that tested in these
experiments, those are throughput testing, delay, jitter, and vCPU Usage [9]. This test is based
on the size of the packet size transmitted by VM A as the sender, to the VM B as the receiver,
which is 64, 128, 256, 384, 512, 640, 768, 896, 1024, 1153, 1280, 1408, and 1514 bytes [9].
These metric value were measured by using D-ITG application. The D-ITG has capable
to producing traffic with various packet level. D-ITG can be used to measure throughput, delay,
jitter across these virtual environment system [9-11]. To ensure the data accuracy, all the test
were execute 20 times, and each run had duration of 10 second [12-13]. Table 1 show virtual
machine specification
Table 1. Virtual Machine Specification
Location Guest OS vCPU RAM Hypervisor vSwitch
Server A VM A
Ubuntu
16.04
1 Core 1 GByte
XenServer
7.0
Open vSwitch
2.5.1
Server B VM B
Ubuntu
16.04
1 Core 1 GByte
XenServer
7.0
Open vSwitch
2.5.1

Bulletin of EEI ISSN: 2302-9285 
Performance Analysis of VXLAN and NVGRE Tunneling... (Mirdan Syahid Mulya Sudrajat)
297
Figure 2. Tunneling system architecture
3. Results and Analysis
The test is conducted to find out which tunneling protocol has the most optimal
throughput value and measure how far the trade-off occurs between VXLAN and NVGRE and
also with native. Based on Figure 3, it can be seen that there is difference of throughput value
between VXLAN and NVGRE performance. The difference occurs when the size of the tested
packet size is 128 -1514 bytes. NVGRE performance has a greater throughput value than
VXLAN performance throughout the packet size testing is increased.
Figure 3. Throughput performance from VM-A to VM-B
The test is conducted to find out which tunneling protocol has the most optimal
throughput value and measure how far the tradeoff occurs between VXLAN and NVGRE and
also with native. Based on Figure 3, it can be seen that there is difference of throughput value
between VXLAN and NVGRE performance. The difference occurs when the size of the tested
packet size is 256 -1514 bytes. NVGRE performance has a greater throughput value than
VXLAN performance throughout the packet size testing is increased. These performance
distance is due the different characteristics of the number of packet header use for each
tunneling platform, the NVGRE has 4 byte GRE header, and a 16 byte VXLAN protocol
consisting of 8 bytes of VXLAN headers and 8 bytes of user datagram protocol (UDP). With less

 ISSN: 2302-9285
298
packet headers, its proven that NVGRE have higher throughput performance than VXLAN. For
the most optimal packet size performance, both of tunneling protocol has the same variable
which is at 1514 bytes, with VXLAN throughput performance of 908.55 Mbps and NVGRE of
916.60 Mbps. It caused of the packet size can impact on the utility of bandwidth usage, so when
the packet size is 1534 bytes, each protocol gets the most optimal throughput.
In Figure 4 shows that the result has the different delay performance. Each tunneling
protocol has a tendency to increase delay value when packet size is tested larger. Tunneling
protocol that has the highest delay performance is VXLAN, it takes on packet size above 768
bytes with delay value 2.98 ms. The maximum performance of delay is not exceeding 3 ms, so
the delay performance in tunneling protocol on virtualization environment is very reliable to use,
because the delay performance accepted with ITU G.1010 standard, related to server access,
which is has <2s standard. This also shows that the performance of Open vSwiitch as a tunnel
end point is able to work efficiently, and the delay performance is not too far away from native
performance. Figure. 5 shows Jitter performance from VM-A to VM-B.
Figure. 4. Delay performances from VM-A to VM-B
Figure. 5. Jitter performance from VM-A to VM-B
There are two sides of the test of vCPU usage performance, at the sender (VM-A) and
recipient side (VM-B). Testing is conducted by sending traffic from VM-A to VM-B by using a
,00
1,00
2,00
3,00
4,00
5,00
128 256 384 512 640 768 896 10241152128014081514
Delay
(ms)
Packet Size (byte)
Delay
Native
NVGRE
VXLAN

Bulletin of EEI ISSN: 2302-9285 
Performance Analysis of VXLAN and NVGRE Tunneling... (Mirdan Syahid Mulya Sudrajat)
299
different variable packet size as shown in Figure 6. For the test results at the sender can be
seen in Figure 6. From the images shown that the VXLAN protocol requires the largest vCPU
usage compared to NVGRE and Native performance. The performance of VXLAN protocol with
NVGRE has the same trend of test result, that is the bigger packet size used then the vCPU
usage performance will decrease. This is due to the greater packet size used, will reduce the
ratio of the number of packet deliveries per second to the bandwidth, so the number of packets
sent per second will decrease.
With the decreasing number of packets per second transmitted, it will be directly decline
the number of packets that must be encapsulated, so the performance of vCPU usage
decreases. At the test result on VM-B can be seen in Figure 7, from the graphic image shows
the vCPU usage in VM-B as a whole has increased performance value compared to VM-A, this
performance improvement shows that vCPU usage during the process of decapsulation in
receiver (VM-B), requires more vCPU usage than the sender's encapsulation process.
Figure 6. vCPU Usage VM-A Figure 7. vCPU Usage VM-B
4. Conclusion
We conclude from the test results and performance analysis of tunneling protocol it
shows that the tunneling protocol, VXLAN and NVGRE has the different performances. The best
tunneling protocol in terms of bandwidth usage efficiency is the NVGRE tunneling protocol, with
an average use of throughput is 771,02 Mbps while VXLAN has 753,62 Mbps. Then the delay
test shows that the tunneling protocol which has the lowest average travel time is the NVGRE
tunneling protocol, with the average delay value of 2.24 ms, while the VXLAN is 2.29 ms. The
second, delay performance of this protocol is in accordance with ITU.TG.1010 related on server
access is <2s. In jitter performance shows that the tunneling protocol which has the lowest
variance delay rate is the VXLAN tunneling protocol with jitter value is 0.348 ms while NVGRE
performance is 0.361 ms. The last is the CPU Usage performance shows that the tunneling
protocol which has the lowest of vCPU usage is the NVGRE tunneling protocol, with the
average CPU Usage value on the sender and recipient host is 60.57% and the VXLAN
performance is 79.84%.
For overall, it can be concluded that the most optimal tunneling protocol performance is
NVGRE because its superiority from 3-of-4 parameters, there are the throughput, vCPU usage,
and delay performance.
References
[1] A Leen, H Franke, C Li, W Liao. Toward Performance Optimization with CPU Offloading for
Virtualized Multi-Tenant Data Center Networks. In IEEE 6th International Conference on Cloud
Computing Technology and Science, 2016.
[2] R Kawashima, H Matsuo. Implementation and Performance Analysis of STT Tunneling using vNIC
Offloading Framework (CVSW). In IEEE 6th International Conference on Cloud Computing
Technology and Science. 2014.s
[3] L Lewin-Eytan, K Barabash, R Cohen, V Jain, A Levin. Designing Modular Overlay Solutions for
Network Virtualization. In IBM Research Report. 2011.

 ISSN: 2302-9285
300
[4] M Sridharanm K Duda, I Ganga, A Greenberg, G Lin, M Pearson, P Thaler, C Tumuluri, N
Venkataramiah, Y Wang. NVGRE: Network Virtualization using Generic Routing Encapsulation.
2011.
[5] Larry L Peterson, Bruce S Davie, Computer Networks, Fifth Edition. Burlington: Elsevier.Inc. 2012
[6] Nadeem, M Aamir Karamat, Taimur. A Survey of Cloud Network Overlay Protocols. In IEEE Sixth
International Conference on Digital Information and Communication Technology and its Applications
(DICTAP), 2016.
[7] Piccolo, V Dell Amamou, Ahmed. Haddadou, Kamel, Pujolle, Guy. A Survey of Network Isolation
Solutions for Multi-tenant data center. In IEEE Communication Survey & Tuttorials. 2015.
[8] Narayan Shaneel, Tauch Sotharith. Network Performance Evaluation of IPv4-v6 Configured Tunnel
and 6to4 Transition Mechanism on Windows Server Operating Systems. In IEEE International
Conference on Computer Design and Applications (ICCDA). 2010.
[9] Alessio Botta, Alberto Dainotti, Antonio Pescape. A tool for the generation of realistic network
workload for emerging networking scenarios. Journal Computer Networks: The International Journal
of Computer and Telecommunications Networking. 2012; 56(15): 3531-3547.
[10] R Kawashima, Ryota Matsuo. Non-Tunneling Edge-Overlay Model Using OpenFlow for Cloud
Datacenter Networks. In IEEE International Conference on Cloud Computing Technology and
Science. 2013.
[11] Perdana, Doan, Riri Fitri Sari. Multicast Routing Performance Analysis for Mobility Models on the
IEEE 1609.4 Standard using Random Dijkstra Algorithm. The 1st International Conference on
Intelligent Green Building and Smart Grid (IGBSG 2014), Taiwan, April 2014.
[12] Munadi, Rendy, et al. Adaptive Control Channel Interval in VANET Based on Mobility Model and
Queuing Network Analysis. Journal of Networks. 2016; 11(2): 62-68. Academic OneFile, Accessed
12 July 2017.
[13] Perdana, Doan, Riri Fitri Sari. Enhancing Channel Coordination Scheme Caused by Corrupted
Nakagami Signal and Mobility Models on the IEEE 1609.4 Standard. JNW 9.12 (2014): 3454-3461.

Performance Analysis of VXLAN and NVGRE Tunneling Protocol on Virtual Network

More Related Content

What's hot (19)

Similar to Performance Analysis of VXLAN and NVGRE Tunneling Protocol on Virtual Network (20)

More from journalBEEI (20)

Recently uploaded (20)

Performance Analysis of VXLAN and NVGRE Tunneling Protocol on Virtual Network