CloudComputing-assignment2

This is the repo imported from TU Berlin GitLab with the code for the assignment 2 in Cloud Computing course.

We got maximum number of points, so enjoy our work.
If you find it useful, you can give us a star.

Tasks:

• Work on 1 host machine
  • Preferably your laptop (physical machine). If you don’t have Linux, use a VM.
• Prepare virtualization environments:
  • Qemu/KVM
  • Docker
• Write 2 new benchmarks:
  • Forksum
  • Nginx
• Execute benchmarks on different virtualization platforms

CPU benchmark questions:

1. Look at your LINPACK measurements. How do they differ between the platforms, what are the main reasons for the differences?

• Native - Avg. 3387000,94 KFLOPS
• KVM - Avg. 3383373,01 KFLOPS
• Qemu - Avg. 49485,31 KFLOPS
• Docker - Avg. 3435431,58 KFLOPS

Benchmarks on native machine, KVM and Docker virtual machines yield similar results. CPU on QEMU without KVM is much slower. This happens, because emulation of the CPU done by QEMU is software based, thus it applies noticeable overhead.

Memory benchmark questions:

1. Look at your memsweep measurements. How do they differ between the platforms, what are the main reasons for the differences?

• Native - Avg. 0,34 s
• KVM - Avg. 0,52 s
• Qemu - Avg. 20,47 s
• Docker - Avg. 0,34 s

In case of Docker, direct access to the MMU is performed => no overhead.

In case of KVM, a dedicated hardware unit is used to translate Guest Physical Address into Host Physical Address => small overhead.

In case of Qemu, shadow page tables are used. Every page fault leads to a so-called VM exit, i.e. switching the contexts from the guest to the host => quite expensive.

Disk benchmark questions:

1. Look at your disk write measurements. How do they differ between the platforms, what are the main reasons for the differences?

• Native - Avg. 787,73 IOPS
• KVM - Avg. 1038,48 IOPS
• Qemu - Avg. 560,60 IOPS
• Docker - Avg. 381,17 IOPS

These results took us by surprise. We suspect that docker performance problems may be related to their overlayfs file system and the fact that docker image was created for each benchmark run - if a new overlay is created on the first write, that would incur significant performance overhead.

2. Which disk format did you use for qemu? How do you expect this benchmark to behave differently on other disk formats?

The disk format used was qcow2. The performance should have been higher if raw disk format was used. For other disk formats, we suspect that the performance would be slightly lower, since qcow2 is qemu's native disk format with the most features and support.

Fork benchmark questions:

1. Look at your fork sum measurements. How do they differ between the platforms, what are the main reasons for the differences?

• Native - Avg. 0,81 s
• KVM - Avg. 1,02 s
• Qemu - Avg. 17,93 s
• Docker - Avg. 1,03 s

In case of Qemu, privileged instructions, needed for a fork, have to be trapped and emulated.

Nginx benchmark questions:

1. Look at your nginx measurements. How do they differ between the platforms, what are the main reasons for the differences?

• Native - Avg. 0,14 s
• KVM - Avg. 40,19 s
• Qemu - Avg. 64,06 s
• Docker - Avg. 0,17 s

As expected native and docker are the fastest. The performance degradation in qemu is easily explained by the choice of networking backend used - User Networking (SLIRP) https://wiki.qemu.org/Documentation/Networking#User_Networking_.28SLIRP.29
This networking backend doesn't require any root privileges and is easiest to use, but adds a lot of overhead.

2. How are these measurements related to the disk benchmark findings?

In our case the measurements do not seem to be related in any significant way.

listing-setup.txt

File containing all commands and inputs you used for installing the tools and preparing your VM image, including comments explaining what the commands do

installing qemu with kvm support

sudo apt install qemu qemu-kvm libvirt-daemon-system libvirt-clients bridge-utils

check if kvm kernel module is enabled

sudo modinfo kvm

download debian wheezy image

curl --output=debian.qcow2 https://people.debian.org/~aurel32/qemu/amd64/debian_wheezy_amd64_standard.qcow2

run debian in qemu

kvm disabled

sudo qemu-system-x86_64 -hda debian.qcow2 -m 1024 -nographic -net user,hostfwd=tcp::2222-:22 -net nic

kvm enabled

sudo qemu-system-x86_64 -enable-kvm -hda debian.qcow2 -m 1024 -nographic -net user,hostfwd=tcp::2222-:22 -net nic

port 2222 on the host is forwarded to 22 (ssh) on the guest

to ssh to the virtual machine

ssh -p 2222 root@localhost # password is root``

guest machine

check internet access in the virtualized os (qemu doesn't support icmp by default, so ping won't work)

dig google.com

setting up the guest system

apt-get update && apt-get upgrade
apt-get install -y fio gcc rsync

Instal nginx

sudo apt-get upgrade
sudo apt-get update
sudo apt-get instal nginx

Create a big file in a localhost localization

This location is different on different systems

cd /usr/share/nginx/html/
sudo fallocate -l 512M big-file.txt

Start nginx

sudo service nginx start

For nginx download benchmarking we are using wrk.
wrk is a benchmarking tool.

Instal wrk

sudo apt-get install build-essential libssl-dev git -y
git clone https://github.com/wg/wrk.git wrk
cd wrk
make

move the executable to somewhere in your PATH, ex:

sudo cp wrk /usr/local/bin

/guest-machine

to synchronize a folder on host and remote machine

rsync -avze "ssh -p 2222" $HOST_PATH root@localhost:$GUEST_PATH # rsync must be installed on both systems

To build image from Dockerfile e.g. for cpu benchmark

docker build -f "cpu.Dockerfile" -t "cpu-benchmark"

To run just created image in docker container

docker run "cpu-benchmark"

To run detached image, with ports mapping between host and container

docker run -d -p 80:80 "nginx-benchmark"