Introduction: Docker and Analysis of its Performance

Sara Vamshi Teja

Research Scholar
University of East Landon, United Kingdom
[email protected]

Abstract
Docker provide some facilities, which are useful for developers and administrators. It is
an open platform can be used for building, distributing, and running applications in a
portable, lightweight runtime and packaging tool, known as Docker Engine. It also provide
Docker Hub, which is a cloud service for sharing applications. Costs can be reduced by
replacing traditional virtual machine with docker container. It excellently reduces the cost
of re-building the cloud development platform.
Keywords: Docker, Docker Container, Virtual Machine, Virtualization, Cloud
Computing.

1. Introduction
Docker is an open source platform that run applications and makes the process easier to
develop, distribute. The applications that are built in the docker are packaged with all the
supporting dependencies into a standard form called a container. These containers keep
running in an isolated way on top of the operating system’s kernel. The extra layer of
abstraction might effect in terms of performance [1].
Even thou, the technologies of the container have been around for over 10 years, but
docker, a generally new hopeful is right now a standout amongst the best innovations, since
it accompanies new capacities that prior technologies did not have. Initially, it gives the
facility to create and control containers. Besides that, applications can easily be packed into
lightweight docker containers by the developer. These virtualized applications can easily be
worked anywhere without any alteration. Moreover, docker can convey more virtual
situations than different innovations, on the same equipment. To wrap things up, docker can
easily coordinate with third-party instruments, which help to easily deploy and manage
docker containers. Docker containers can easily be deployed into the cloudbased
environment [2].
This paper is a review on technology of docker, and will analyse its performance by a
systematic literature review. The article is organised as follow. Next section will introduce
the technology of docker. In Section 3, a more detailed description of docker and its
components will be presented. Section 4 briefly compare technology of Virtual Machine
and Docker. Sections 5 and 6 will discuss the advantages and disadvantages of docker
container, respectively. In Section 6 and 7, I briefly review few recent researches on
measuring the performance of Docker and compare it with other container technologies.
Finally, in section 9 and 10, features in virtual machines and containers will be briefly
summarised, following with a short summary of the paper.
1.1 What is Docker?
Docker provides a facility to automate the applications when they are deployed into
Containers. In a Container environment where the applications are virtualized and executed,
docker adds up an extra layer of deployment engine on top of it. The way that docker is
designed is to give a quick and a lightweight environment where code can be run efficiently
and moreover it provides an extra facility of the proficient work process to take the code
from the computer for testing before production [9]. Russell (2015) confirms that, as quick
as it is possible docker allows you to test your code and deploy it into the production
environment [6]. Turnbull (2014) concludes by saying that docker is amazingly simple [9].
Certainly, you can begin with a docker with a simple configuration system, a docker binary
with Linux kernel.

1.2 What is a Docker container and how is it different than a VM?  Does
containerization replace my virtualization infrastructure?
Containerization is very different from virtualization. It starts with the Docker engine, the
tool that creates and runs containers (1 or more), and is the Docker installed software on
any physical, virtual or cloud host with a compatible OS. Containerization leverages the
kernel within the host operating system to run multiple root file systems. We call these root
file systems “containers.” Each container shares the kernel within the host OS, allowing
you to run multiple Docker containers on the same host. Unlike VMs, containers do not
have an OS within it. They simply share the underlying kernel with the other containers.
Each container running on a host is completely isolated so applications running on the same
host are unaware of each other (you can use Docker Networking to create a multi-host
overlay network that enables containers running on hosts to speak to one another).

1.3 From an infrastructure standpoint, what do I need from Docker? Is Docker a

piece of hardware running in my datacenter, and how taxing is it on my environment?
The Docker engine is the software that is installed on the host (bare metal server, VM or
public cloud instance) and is the only “Docker infrastructure” you’ll need. The tool creates,
runs and manages Docker containers. So actually, there is no hardware installation
necessary at all.
The Docker Engine itself is very lightweight, weighing in around 80 MB total.

1.4 How does Docker help manage my infrastructure? Do I containerize all my

infrastructure or something?
Docker isn’t focused on managing your infrastructure. The platform, which is
infrastructure agnostic, manages your applications and helps ensure that they can run
smoothly, regardless of infrastructure type via solutions like Docker Datacenter. This gives
your company the agility, portability and control you require. Your team is responsible for
managing the actual infrastructure.

1.5 How many containers can run per host?

As far as the number of containers that can be run, this really depends on your
environment. The size of your applications as well as the amount of available resources (i.e
like CPU) will all affect the number of containers that can be run in your environment.
Containers unfortunately are not magical. They can’t create new CPU from scratch. They
do, however, provide a more efficient way of utilizing your resources. The containers
themselves are super lightweight (remember, shared OS vs individual OS per container)
and only last as long as the process they are running. Immutable infrastructure if you will.
2. Docker Inside
There are four main internal components of docker, including Docker Client and Server,
Docker Images, Docker Registries, and Docker Containers. These components will be
explained in details in the following sections.

2.1 Docker Client and Server

Docker can be explained as a client and server based application, as depicted in Figure 1.
The docker server gets the request from the docker client and then process it accordingly.
The complete RESTful (Representational state transfer) API and a command line client
binary are shipped by docker. Docker daemon/server and docker client can be run on the
same machine or a local docker client can be connected with a remote server or daemon,
which is running on another machine [9].

Figure. 1 Docker architecture.

2.2 Docker Images
There are two methods to build an image. The first one is to build an image by using a
read-only template. The foundation of every image is a base image. Operating system
images are basically the base images, such as Ubuntu 14.04 LTS, or Fedora 20. The images
of operating system create a container with an ability of complete running OS. Base image
can also be created from the scratch. Required applications can be added to the base image
by modifying it, but it is necessary to build a new image. The process of building a new
image is called “committing a change”. The second method is to create a docker file. The
docker file contains a list of instructions when “Docker build” command is run from the
bash terminal it follows all the instructions given in the docker file and builds an image.
This is an automated way of building an image.
2.3 Docker Registries
Docker images are placed in docker registries. It works correspondingly to source code
repositories where images can be pushed or pulled from a single source. There are two
types of registries, public and private. Docker Hub is called a public registry where
everyone can pull available images and push their own images without creating an image
from the scratch. Images can be distributed to a particular area (public or private) by using
docker hub feature.

2.4 Docker Containers

Docker image creates a docker container. Containers hold the whole kit required for an
application, so the application can be run in an isolated way. For example, suppose there is
an image of Ubuntu OS with SQL SERVER, when this image is run with docker run
command, then a container will be created and SQL SERVER will be running on Ubuntu
OS.

3. Virtual Machine vs. Docker

Virtualization is an old concept, which has been in used in cloud computing, after IaaS has
been accepted as a crucial technique for system constitution, resource provisioning, and
multi-tenancy. Virtualized resources play the main role in solving the problems using the
core technique of cloud computing. The Figure 2 shows the architecture of the virtual
machine.

Figure. 2 Virtual Machine architecture.

Hypervisor is lying between host and guest operating systems. It is a virtual platform and it
handles more than one operating system in the server. It works between the operating
system and CPU. The virtualization divides it into two segments: the first one is Para-
Virtualization and the second one is Full Virtualization [3]. Figure 3 depicts the architecture
of the Docker Container.
Linux containers are managed by the docker tool and it is used as a method of operating
system level virtualization. Figure 3 shows that in single control host there are many Linux
containers, which are isolated. Resources such as Network, Memory, CPU, and Block I/O
are allocated by Linux kernel and it also deals with cgroups without starting virtualization
machine [8].

Figure 3 Docker Container architecture.

According to Waldspurger (2002), in the Linux containers, an architecture is to manage

CPU and distribute its resources more proficiently. In any example of Hyper-V or
VMWare, because of overhead incurred, it is not easy to run more than ten virtual machines
[13]. Up to a great extent, this issue has been solved by the containers. Containers only
utilize those resources, which are needed for the services or applications. Therefore, on a
weak configured machine, above 50 requests of the containers can be executed.
For example, suppose an organisation provides email security services. The major
functions of these services are to check emails for viruses, spam, and malware. Moreover, it
could manage to transfer messages to the agent, logs and report delivery failure if the
product is installed in the cloud [10]. Mostly in these cases, there is no use of any
associated dependencies or OS level libraries or any kernel data structure. Therefore, it is
worthwhile to containerized every component by sandboxing them utilizing OpenVZ or
Docker instead of having virtual machines.
In many enterprises, virtual machines are used to perform element testing. In this process,
a lot of CPU resources and memory space are consumed. Whereas, container technology
provides a guarantee to their users that excess of a workload would not affect the efficiency
of the resources. The container takes less time for installation as compared to virtual
machines, so the adaptability of containers is much higher than VMs.
Furthermore, both Docker and OpenVZ have been under great examination in terms of
their security aspects. When isolation is reduced, it directly affects the security, which also
decreases rapidly. Root users of Linux can easily get access to containers as containers also
use the same kernel and operating system. The isolation of docker is not as strong as a
virtual machine, even though docker isolates the application, which is running in the docker
container from its primary host. Additionally, it is possible that some of the applications
would not be able to run in a containerized technology and they need to run on a different
operating system.

4. Advantages of Docker Container

The demand and the advancement of Linux containers can be seen in the last few years.
Docker has become popular very quickly, because of the benefits provided by docker
container. The main advantages of docker are speed, portability, scalability, rapid delivery,
and density.

4.1 Speed
Speed is one of the most exceedingly touted advantages of Containers. When the benefits
of using docker are highlighted, it would be incredible not to mention about the speed of
docker in the conversation (Chavis & Architect, 2015). The time required to build a
container is very fast because they are really small. Development, testing, and deployment
can be done faster as containers are small. Containers can be pushed for testing once they
have been built and then from there, on to the production environment [12].

4.2 Portability
Those applications that are built inside docker containers are extremely portable. These
portable applications can easily be moved as a single element and the performance remains
the same [12].

4.3 Scalability
Docker has the ability that it can be deployed in several physical servers, data servers, and
cloud platforms. It can also be run on every Linux machine. Containers can easily be
moved from a cloud environment to local host and from there back to cloud again at a fast
pace. Adjustments can easily be done; the scale can simply be adjusted by the user
according to the need [5].

4.4 Rapid Delivery

The format of a Docker Containers is standardized so programmers do not have to stress
over one another’s tasks. The responsibility of the administrator is to deploy and maintain
the server with containers, whereas the responsibility of the programme is to look after the
applications inside the docker container. Containers can work in every environment as they
have all the required dependencies embedded within the applications and they are all tested
[12]. Docker provides a reliable, consistent, and improved environment, so predictable
results can be achieved when codes are moved between development, test and production
systems (Chavis & Architect, 2015).

4.5 Density
Docker uses the resources that are available more efficiently because it does not use a
hypervisor. This is the reason that more containers can be run on a single host as compared
to virtual machines. The performance of a Docker Containers is higher because of higher
density and no overhead wastage of resources [5].

5. Disadvantages of Docker Container

There are some drawbacks of docker containers, which are listed below [1, 4]:
 Complete virtualization is not provided by a docker because it depends on the Linux
kernel, which is provided by the local host.
 Currently, docker does not run on older machines. It only supports 64-bit local
machines.
 The complete virtualized environment must be provided by the docker container for
Windows and Mac machines. Even though the boot2docker tool fills this gap, but still, it
should be checked whether it makes obstructions to acceptance by users of these systems or
the integration and performance with the host machine’s operating system are adequate [4].
 It is necessary that the possibility of security issues should be evaluated. Building
off trusting binaries could be made easier by digitally signing docker images, for future
support.
 An important concern is to check if the teaching community or scientific researcher
will significantly think of adopting docker.

6. Docker Performance
Seo et al. (2014) used two servers with the same configuration in the cloud environment.
One server was used for docker and the other one was for an Open Stack platform for KVM
by means of a virtualization tool [8]. According to him, a VM works independently. This
factor make it easy to apply and manage the policy of network, security, user, and the
system. However, docker does not contain a guest operating system. Therefore, it takes
very little time in distributing and gathering images. The boot time is also very short. These
are the main advantages of utilizing Docker Cloud as compared with VM Cloud. Scheepers
(2014) compares LXC and Xen virtualization technologies to benchmark some applications
[7]. He explains that Xen would be a better choice in the sense of equally distributing
resources, performance is not dependent on the other tasks, and it is executed on the same
machine. However, LXC is much better in the sense of getting most of the hardware
resources or for the execution of smaller isolated processes. In private and dot clouds, LXC
is a better option.
Felter et al. (2014) evaluate the performance of three different environments, Native,
Docker, and KVM [3]. He clarifies that containers and VMs are both mature innovation that
has profited from last 10 years of incremental equipment and programming enhancements.
According to this research, docker is equivalent to or surpasses KVM execution for each
situation they tried. Their outcomes demonstrate that both KVM and docker present
irrelevant overhead for CPU and memory execution. It has also been shown that the overall
performance of docker is better than the Local Host, as the applications were executed and
responded faster than in Local Host. Moreover, fewer hardware resources were used in
docker container to perform the tasks.
 Docker is really a future demanding technology. As users and developers would know
more about the docker and its capabilities then they would consider replacing traditional
virtualization with docker technology. Docker provides many simple and useful features.
To get the best performance and results, it is highly recommended to move up from the
default configuration. Containers provide advanced density, better performance, scalability,
and usability as compared with traditional virtualization because containers smartly utilize
its resources, which reduce the chance of unnecessary overhead. Containers are better in
performance than virtual machine, because containers take less start-up time. Docker has
removed the biggest issue of “dependency”. Now containers have all of their required
dependencies, which help containers to be properly built, and to execute them in any docker
environment. An additional layer of isolation is provided by the container, which increases
the containers’ security. Docker is not as insecure as people normally think, but it provides
a complete protection.

7. Docker vs. other Container Technology

In this section, the performance of application virtualization and the performance of the
docker container will be discussed, and the evaluation of other containerize technology will
be compared and reviewed. Seo et al. (2014) summarize that there is no guest OS of docker
in the cloud, so the storage and the wastage of CPU resources are less [8]. The images are
not disturbed; boot time is faster and the time of generating the images is short. These are
the benefits of docker cloud in comparison with VM Cloud. They used two similar servers
with the same configuration in the cloud environment. One server was used for docker and
the other one was for an Open Stack platform for KVM by means of a virtualization tool.
Ubuntu Server was used as a base platform [8].
To calculate the approximate boot-time, 20 images were generated on each server and boot
time was checked. Figure 4 shows that the boot time of docker is lesser than the boot time
of KVM. Docker uses the Host OS, whereas KVM uses Guest OS. Thus, the boot time of
docker is shorter than the boot time of KVM.

Figure. 4 Docker vs KVM Average boot time.

To calculate the operational speed, python language was used. Figure 5 shows that
operation speed of 100,000 is averagely around 4.5s. To measure the operation speed, they
obtain the average process time and standard deviation, by repeating the same process 100
times on docker and VM.
 Figure. 5 CPU Calculation Performance.

Figure 5 shows the calculation speed of docker is slightly faster than the calculation speed
of the VM [7]. Seo et al. (2014) concluded that VM works independently [8]. This is one of
the reasons that it is easy to apply and manage the policy of network, security, user, and the
system. However, docker does not contain a guest operating System. Therefore, it takes
very less time in distributing and gathering images. Its boot time is also very short. These
are the main advantages of utilizing docker cloud as compared with VM Cloud.
Scheepers (2014) compares LXC and Xen virtualization technologies to benchmark some
applications [7]. For this purpose, Scheepers uses two servers Core OS 324.3.0 and
XenServer 6.2 with docker version 0.11.1. The configuration of these systems is RAM
4GB, CPU Intel Xeon Quad core and the virtualization support is Intel VTX. The base
operating system is Ubuntu 12.04 and containers will run on both machines. 2GB of
memory is allotted to the first virtual machine and Apache 2.2, WordPress 3.9 and PHP 5.3.
This was used as an application Server. 1GB of memory is used by the second virtual
machine with MYSQL Database 5.5. This database was filled by the WordPress sample
contents. This machine was used as a database Server. JMeter was used as a benchmarked
tool.
Figure 6 shows that LXC experienced less overhead as compared to Xen when the
SELECT query was run. The focus on running this benchmark process is to see the
utilization of the CPU and the performance of the Network speed because these are the
main resources consumed in this test.

Figure. 6 Time in millisecond to complete a one SQL select query.

 Figure 7 shows that in Xen setup, it took 16 seconds to accomplish when the INSERT
query was run in the database, whereas in LXC setup it took longer–around 335 seconds.
This reveals the inability of LXC container to isolate resources efficiently.

Figure. 7 Time in millisecond to complete 10,000 SQL INSERT queries.

Scheepers (2014) concludes that Xen would be a better choice in the sense of equally
distributing resources, performance is not dependent on the other tasks, and it is executed
on the same machine [7]. However, LXC is much better in the sense of utilizing most of the
hardware resources or the execution of smaller isolated processes. In private and dot clouds,
LXC is a better option. Felter et al. (2014) evaluated the performance of three different
environments, Native, Docker, and KVM [3]. Overhead issues are also highlighted in their
research. Scenarios were investigated where more than one hardware resource was
completely utilized. To perform the tests, they used an IBM x3650 M4 server, 16 core
processors of Xeon E5-2665, Two Intel Sandy Bridge-EP of 2.4 - 3.0 GHz and 256 GB of
RAM. To make a non-uniform memory access, two processors were linked together with
QPI link. Cloud providers also use this kind of similar Server. The base operating system
was Ubuntu 13.10, docker version 1.0, Linux kernel 3.11.0, libvirt version 1.1.1 and QEMU
1.5.0. This Figure 8 shows that the average size of 1 MB was used for I/O, little over 60
seconds by measuring the performance of sequential read and write. In this case, slight
overhead can be seen by Docker and KVM. In other cases, KVM has almost a four times
performance difference.

Figure. 8 Sequential I/O throughput (MB/sec).

Figure 9 demonstrates the execution of irregularly read, write and mixed workloads
utilizing a 4 kB square size and simultaneousness of 128, which I tentatively decided gives
the greatest execution to this specific SSD. As I would expect, docker acquaints no
overhead contrasted and Linux, however, KVM conveys just have the same number of
IOPS since every I/O operation must experience QEMU. While the VM's supreme
execution is still very high, it utilizes more CPU cycles per I/O operation, leaving less CPU
accessible for application work.

Figure. 9 Random I/O throughput (IOPS).

Felter et al. (2014) conclude that containers and VMs are both mature innovations that
have profited from the last 10 years of incremental equipment and programming
enhancements [3]. When all is said and done, docker is equivalent to or surpasses KVM
execution for each situation I tried. Our outcomes demonstrate that both KVM and docker
present irrelevant overhead for CPU and memory execution.
To conclude these past works, regardless of utilizing distinctive techniques and having
diverse centres, one thing is common.

8. Virtual Machines vs. Containers

Table 1 compare features of different containerized and virtual machine technologies.
Virtual machine uses an extra layer between the host operating system and guest operating
system. This layer is known as a Hypervisor. Whereas docker adds up an extra layer
between host operating systems and where the applications are virtualized and executed,
which is known as a Docker Engine. As docker does not use any guest operating system
that makes a big difference in performance between a docker container and a virtual
machine technology. In Table 1, the performances of applications running in different
containers and virtual machines are also briefly compared.
As it is given in the table above, according to Seo et al. (2014) the docker performance is
better than KVM, in terms of boot time and calculation speed [8], whereas Felter et al.
(2014) proves that there is no difference of wastage of resources (overhead) between
Docker and KVM but there is a noticeable difference in execution, as KVM is faster than
Docker [3]. Scheepers (2014) found out that LXC takes a longer time to accomplish tasks,
whereas Xen Server takes less time [7]. LXC is better in the sense of fewer wasted
resources while Xen is better in the sense of equally distributing resources.

Table 1: Comparison Table based on Different Virtual Machines and Containerized

Technology
 Seo et al. (2014) [8] Scheepers (2014) [7] Felter et al. (2014) [3]
XenServer CoreOS
Docker KVM (Xen) (LXC) Native Docker KVM
More Less Overhead Slightly Slightly less
Boot Time Boot time overhead overhead (wastage of less than
(wastage of (wastage of resources) overhead Native and
short long than
resources) resources) Docker
Native
Less time to Longer time Slow Slow
Calculation Calculation
accomplish to Execution Execution Fast
Speed is Speed is
request accomplish equal to equal to Execution
faster Slower
request Docker Native
Better in Better in
sense of sense of
No Guest Works equally Mature Mature
executing -
OS Independently distributing Innovation Innovation
isolated
resources
processes

9. Conclusion:
Docker automates the applications when they are containerized. An extra layer of docker
engine is added to the host operating system. The performance of docker is faster than
virtual machines as it has no guest operating system and less resource overhead.

References
[1] Boettiger, C. (2015). An introduction to Docker for reproducible research. ACM
SIGOPS Operating Systems Review, 49(1), 71-79.
[2] Bui, T. (2015). Analysis of docker security. arXiv preprint arXiv:1501.02967.
[3] Felter, W., Ferreira, A., Rajamony, R., & Rubio, J. (2014). An updated performance
comparison of virtual machines and linux containers. technology, 28, 32.
[4] Harji, A. S., Buhr, P. A., & Brecht, T. (2013). Our troubles with Linux Kernel
upgrades and why you should care. ACM SIGOPS Operating Systems Review,
47(2), 66-72.
[5] Joy, A. M. (2015). Performance comparison between Linux containers and virtual
machines. Paper presented at the Computer Engineering and Applications
(ICACEA), 2015 International Conference on Advances in.
[6] Russell, B. (2015). Passive Benchmarking with docker LXC, KVM & OpenStack.
[7] Scheepers, M. J. (2014). Virtualization and containerization of application
infrastructure: A comparison.
[8] Seo, K.-T., Hwang, H.-S., Moon, I.-Y., Kwon, O.-Y., & Kim, B.-J. (2014).
Performance Comparison Analysis of Linux Container and Virtual Machine for
Building Cloud.
[9] Turnbull, J. (2014). The Docker Book: Containerization is the new virtualization.
[10] Van der Aalst, W., Weijters, T., & Maruster, L. (2004). Workflow mining:
Discovering process models from event logs. Knowledge and Data Engineering,
IEEE Transactions on, 16(9), 1128-1142.
[11] Varghese, B., Subba, L. T., Thai, L., & Barker, A. (2016). Container-Based Cloud
Virtual Machine Benchmarking. arXiv preprint arXiv:1601.03872.
[12] Vase, T. (2015). Advantages of Docker.
[13] Waldspurger, C. A. (2002). Memory resource management in VMware ESX server.
ACM SIGOPS Operating Systems Review, 36(SI), 181-194.
[14] ACM SIGOPS Operating Systems Review. 36 p.181-194.