Characterizing Service Level Objectives for Cloud Services: Realities ...

[Pages:26]Characterizing Service Level Objectives for Cloud Services: Realities and Myths

. Ding, R. Cao, I. Saravanan, N. Morris and C. Stewart, 2019 IEEE International Conference on Auto Computing (ICAC), Umea, Sweden, 2019, pp. 200-206.

Summarized by: Ibifubara Iganibo

line

! Introduction ! Motivation ! Service Level Objectives ! Service Level Objective Structures ! Systematic Literature Review ! Characterizing Realities and Myths ! Conclusion

oduction

! Service level objectives (SLOs) stipulate performance goals for clou applications, microservices, and infrastructure.

! SLOs are widely used because system managers can tailor goals to products, companies, and workloads.

! SLOs outline quality of service goals and service providers must sati these goals or risk penalties.

! SLOs allow cloud applications to safely embed content from third p

tivation

! Performance oriented Service level objectives have been widely adopted over the past 15 years.

tivation

! Some textbooks discuss the structure of SLOs including performance goals:

! But only sketch SLO goals and use outdated examples

! Performance goals are an emerging trend and details change often.

! So, textbook authors will struggle to keep content up to date, especially as new workloads and expectations emerge

al of the er

! This paper studied real SLOs, quantified their performance goals and labeled common perceptions about their goals as realities or myths.

! Systematic literature review (SLR) to prune results.

ntributions

! They collected diverse and realistic cloud SLOs using systematic literature review.

! Grouped and compared SLOs by source, target workload and performance goals.

! Their analysis confirms common assumptions and provides reason to reconsider others.

umptions

Dataset does corroborate 3 common assumptions.

! SLO delays differ greatly across workloads

! 90th, 95th and 99th percentiles are widely used SLO percentile goals

! Strict SLOs targeting single digit delays are uncommon in practice

Dataset does not corroborate the following assumptions

! Minute granularity reporting periods are common

! Response time goals used in academic papers mirror goals set in practice.

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download