TY - GEN
T1 - Reducing the Tail Latency of Microservices Applications via Optimal Configuration Tuning
AU - Somashekar, G.
AU - Suresh, A.
AU - Tyagi, S.
AU - Dhyani, V.
AU - Donkada, K.
AU - Pradhan, A.
AU - Gandhi, A.
N1 - Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - The microservice architecture is an architectural style for designing applications that supports a collection of fine-grained and loosely-coupled services, called microservices, enabling independent development and deployment. An undesirable complexity that results from this style is the large state space of possibly inter-dependent configuration parameters (of the constituent microservices) which have to be tuned to improve application performance.This paper investigates optimization algorithms to address the problem of configuration tuning of microservices applications. To address the critical issue of large state space, practical dimensionality reduction strategies are developed based on available system characteristics. The evaluation of the optimization algorithms and dimensionality reduction techniques across three popular benchmarking applications highlights the importance of configuration tuning to reduce tail latency (by as much as 46%). A detailed analysis of the efficacy of different dimensionality reduction techniques in capturing the most important parameters is performed using ANOVA techniques. Results show that the right combination of optimization algorithms and dimensionality reduction can provide substantial latency improvements by identifying the right subset of parameters to tune, reducing the search space by as much as 83%.
AB - The microservice architecture is an architectural style for designing applications that supports a collection of fine-grained and loosely-coupled services, called microservices, enabling independent development and deployment. An undesirable complexity that results from this style is the large state space of possibly inter-dependent configuration parameters (of the constituent microservices) which have to be tuned to improve application performance.This paper investigates optimization algorithms to address the problem of configuration tuning of microservices applications. To address the critical issue of large state space, practical dimensionality reduction strategies are developed based on available system characteristics. The evaluation of the optimization algorithms and dimensionality reduction techniques across three popular benchmarking applications highlights the importance of configuration tuning to reduce tail latency (by as much as 46%). A detailed analysis of the efficacy of different dimensionality reduction techniques in capturing the most important parameters is performed using ANOVA techniques. Results show that the right combination of optimization algorithms and dimensionality reduction can provide substantial latency improvements by identifying the right subset of parameters to tune, reducing the search space by as much as 83%.
KW - configuration tuning
KW - dimensionality reduction
KW - microservices
KW - ML for systems
KW - optimization
KW - tail latency
UR - https://www.scopus.com/pages/publications/85142367924
U2 - 10.1109/ACSOS55765.2022.00029
DO - 10.1109/ACSOS55765.2022.00029
M3 - Conference contribution
AN - SCOPUS:85142367924
T3 - Proceedings - 2022 IEEE International Conference on Autonomic Computing and Self-Organizing Systems, ACSOS 2022
SP - 111
EP - 120
BT - Proceedings - 2022 IEEE International Conference on Autonomic Computing and Self-Organizing Systems, ACSOS 2022
A2 - Casadei, Roberto
A2 - Di Nitto, Elisabetta
A2 - Gerostathopoulos, Ilias
A2 - Pianini, Danilo
A2 - Dusparic, Ivana
A2 - Wood, Timothy
A2 - Nelson, Phyllis
A2 - Pournaras, Evangelos
A2 - Bencomo, Nelly
A2 - Gotz, Sebastian
A2 - Krupitzer, Christian
A2 - Raibulet, Claudia
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 3rd IEEE International Conference on Autonomic Computing and Self-Organizing Systems, ACSOS 2022
Y2 - 19 September 2022 through 23 September 2022
ER -