Digitala Vetenskapliga Arkivet

Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Experimental Investigation of Container-based Virtualization Platforms For a Cassandra Cluster
2017 (English)Independent thesis Advanced level (professional degree), 20 credits / 30 HE creditsStudent thesis
Abstract [en]

Context. Cloud computing is growing fast and has established itself as the next generationsoftware infrastructure. A major role in cloud computing is the virtualization of hardware toisolate systems from each other. This virtualization is often done with Virtual Machines thatemulate both hardware and software, which in turn makes the process isolation expensive. Newtechniques, known as Microservices or containers, has been developed to deal with the overhead.The infrastructure is conjoint with storing, processing and serving vast and unstructureddata sets. The overall cloud system needs to have high performance while providing scalabilityand easy deployment. Microservices can be introduced for all kinds of applications in a cloudcomputing network, and be a better fit for certain products.Objectives. In this study we investigate how a small system consisting of a Cassandra clusterperform while encapsulated in LXC and Docker containers, compared to a non virtualizedstructure. A specific loader is built to stress the cluster to find the limits of the containers.Methods. We constructed an experiment on a three node Cassandra cluster. Test data is sentfrom the Cassandra-loader from another server in the network. The Cassandra processes are thendeployed in the different architectures and tested. During these tests the metrics CPU, disk I/O,network I/O are monitored on the four servers. The data from the metrics is used in statisticalanalysis to find significant deviations.Results. Three experiments are being conducted and monitored. The Cluster test pointed outthat isolated Docker container indicate major latency during disk reads. A local stress test furtherconfirmed those results. The step-wise test in turn, implied that disk read latencies happened dueto isolated Docker containers needs to read more data to handle these requests. All Microservicesprovide some overheads, but fall behind the most for read requests.Conclusions. The results in this study show that virtualization of Cassandra nodes in a clusterbring latency in comparison to a non virtualized solution for write operations. However, thoselatencies can be neglected if scalability in a system is the main focus. For read operationsall microservices had reduced performance and isolated Docker containers brought out thehighest overhead. This is due to the file system used in those containers, which makes disk I/Oslower compared to the other structures. If a Cassandra cluster is to be launched in a containerenvironment we recommend a Docker container with mounted disks to bypass Dockers filesystem or a LXC solution.

Place, publisher, year, edition, pages
2017. , p. 109
Keywords [en]
Container Virtualization, Cassandra, Docker, LXC, Big data, Microservices, Linux distributions
National Category
Computer Systems
Identifiers
URN: urn:nbn:se:bth-14544OAI: oai:DiVA.org:bth-14544DiVA, id: diva2:1110959
External cooperation
Qvantel Sweden AB
Subject / course
Degree Project in Master of Science in Engineering 30.0
Educational program
CIDAY Master of Science in Computer and Electrical Engineering
Presentation
2017-05-30, 17:09 (Swedish)
Supervisors
Examiners
Available from: 2017-06-21 Created: 2017-06-16 Last updated: 2022-05-12Bibliographically approved

Open Access in DiVA

fulltext(2820 kB)933 downloads
File information
File name FULLTEXT02.pdfFile size 2820 kBChecksum SHA-512
2eed778afac32e787dbd694631b43f5ed990f15652b69c9e8478d636be93026bea4ef3f2ae65b9bc250f1badd3bb26a5384d0eeabba2273b6a33b863c6253dcf
Type fulltextMimetype application/pdf

Search in DiVA

By author/editor
Sulewski, Patryk
Computer Systems

Search outside of DiVA

GoogleGoogle Scholar
Total: 933 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 961 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf