- Title Information
- Title
- Stability of Parallel-Server Systems with Service Elasticity
- Abstract
- Large-scale cloud networks and data centers operate in distributed parallel queue systems, rather than maintaining a centralized queue. Since demand patterns vary depending on time and cannot be predicted with certainty, large-scale cloud networks and data centers face the serious issue of achieving efficient server utilization and energy consumption whilst minimizing user-perceived delay. Achieving optimal service elasticity especially without global queue length information is very challenging - turning off idle servers saves energy, but turning them on when needed requires a long setup period, which typically exceeds the latency tolerance of real-time processing and control functions by orders-of-magnitude.
Recently, a token-based joint auto-scaling and load balancing strategy that maintains constant communication overhead has been proposed in the literature for such systems. The scheme achieves certain asymptotic optimality in terms of delay performance and energy consumption when the number of servers, N, tends to infinity. However, for a fixed finite value of N, it is not clear that the system under the proposed scheme is even stable. In fact, for some choice of parameters it is known that the system is unstable for small N. Our primary goal in this project is to investigate this stability issue for finite N-values.
Specifically, we propose an alternative scheme that provides better service elasticity and robustness for systems with finite N, whilst still maintain- ing constant communication overhead. We prove the stability of the pro- posed scheme and corroborate these theoretical results using simulations. Using additional simulations, we also investigate the scheme’s performance in terms of average expected wait time and average expected energy consumption compared to the existing token-based joint auto-scaling and load balancing scheme. The results show that the proposed scheme performs better in both aspects.
- Name:
Personal
- Name Part
- Sohan, Misha Wei
- Role
- Role Term (marcrelator)
(authorityURI="http://id.loc.gov/vocabulary/relators", valueURI="http://id.loc.gov/vocabulary/relators/cre")
- creator
- Name:
Personal
- Name Part
- Ramanan, Kavita
- Role
- Role Term (marcrelator)
(authorityURI="http://id.loc.gov/vocabulary/relators", valueURI="http://id.loc.gov/vocabulary/relators/ths")
- thesis advisor
- Name:
Personal
- Name Part
- Harrison, Matthew
- Role
- Role Term
- reader
- Name:
Corporate
- Name Part
- Brown University. Applied Mathematics
- Role
- Role Term:
Text
- sponsor
- Origin Information
- Copyright Date
- 2019
- Type of Resource
- text
- Physical Description
- digitalOrigin
- born digital
- Language
- Language Term:
Text (ISO639-2B)
(authorityURI="http://id.loc.gov/vocabulary/iso639-2.html", valueURI="http://id.loc.gov/vocabulary/iso639-2/eng")
- English
- Note:
thesis
- Senior thesis (ScB)--Brown University, 2019
- Note
(displayLabel="Concentration")
- Applied Mathematics, Economics, and Computer Science
- Genre (aat)
- theses
- Subject (fast)
(authorityURI="http://id.worldcat.org/fast", valueURI="http://id.worldcat.org/fast/01085713")
- Topic
- Queuing theory
- Identifier:
DOI
- 10.26300/28ha-4723
- Access Condition:
rights statement
(href="http://rightsstatements.org/vocab/InC/1.0/")
- In Copyright
- Access Condition:
restriction on access
- Collection is open for research.