Diese Website verwendet Cookies

Auf unserer Website verwenden wir Cookies, die für die Darstellung der Website zwingend erforderlich sind. Mit Klick auf „Auswahl akzeptieren“ werden nur diese Cookies gesetzt. Andere Cookies und Inhalte von Drittanbietern (z.B. YouTube Videos oder Google Analytics), werden nur mit Ihrer Zustimmung durch Auswahl oder Klick auf „Alles akzeptieren“ gesetzt. Weitere Einzelheiten erfahren Sie in unserer Datenschutzerklärung, in der Sie Ihre Einwilligung jederzeit widerrufen können.

02.11.2021

benchANT Benchmarking Insight

AWS EC2 vs Open Telekom Cloud vs IONOS Cloud

This paper discusses the price-performance trade-offs when selecting a cloud provider for a specific database set-up. More precisely, we investigate how a 3-node MongoDB cluster
performs on various virtual machine flavors of different cloud providers using a write-heavy workload. All flavors we use, run with 2 virtual cores (as selected in the cloud provider API)
and make use of 8 GB of memory. Overall, we picked 6 flavors from the German cloud provider IONOS, 4 flavors from the German cloud provider Open Telekom Cloud (OTC), and 4 flavors from market leader Amazon Web Services/EC2. We apply the widely used Yahoo Cloud Serving Benchmark (YCSB) to fill the database initially with 2.000.000 items of 5kBit items size. Our YCSB workload configuration executes 10.000.000 operations with 80% writes and 20% reads as it can be found in many IoT scenarios.

Using our benchmarking platform benchANT, we identified several surprising and counterintuitive results that show that benchmarking is necessary in advance to selecting a cloud
provider and a cloud flavor type. The key insights of our research are the following:

  • Costs are varying tremendously between flavors and offerings. The cheapest setup is around 170 € while the most expensive one is above 360 €.
  • The smaller cloud providers can compete with AWS in terms of throughput, read latency, and write latency. In terms of cost the well performing solutions are more expensive than the AWS configurations (worse performing). Yet, considering price for value, the two small providers top out higher than AWS.
  • AWS has an astonishing stability across all flavors when it comes to value for money. The other two cloud providers are a lot more versatile over their flavor spectrum.
  • For write-heavy workloads MongoDB does not perform worse when run on HDD instead of SSD when journaling is disabled. Indeed, there may even be performance benefits when using an HDD. This could be a consequence of MongoDB’s internal workings, but may also be caused by the fact that I/O bandwidth is limited by some cloud providers: The bandwidth for SSDs growths with the size of the storage volume, while HDDs have a fixed bandwidth.
  • There is no one size fits all flavor performing good for all KPIs under investigation. For instance, flavors with excellent write latency are not necessarily the ones performing good for the read latency; and expensive flavors are not always high-performance flavors. In particular, a flavor running HDD is amongst the cheapest, but still the best when it comes to write latency. Yet, it is the by far worst, with respect to read latency. 

These results indicate clearly that there is a significant need for reliable, easily accessible, and comprehensive data on performance and other KPIs of different database-cloud configurations on a per workload bases. The right choice can save up to 190 € (50%) per month for a three-node cluster or increase measured performance by a factor of three.

benchANT offers a reliable, easy-to-use, Web-based self-service platform to evaluate beforehand and helping you to make the right decision.

Yasmin Abu Dorrah

Ansprechpartnerin

Yasmin Abu Dorrah

PR und Kommunikationsmanagerin

E-Mail
AbuDorrah@bwcon.de
Telefon
+49 (0) 176 45255676