Interleaved Tenant Placement in Database-as-a-Service
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing database management systems in cloud environments face challenges in efficiently allocating and migrating tenants across servers while maintaining service-level objectives (SLOs) and minimizing operational costs, particularly due to unpredictable load patterns and the need for dynamic resource adjustments.
Innovation Solution
A method for dynamically measuring workload and applying an adaptive migration command that interleaves tenant replication across servers, minimizing the number of active servers and maximizing performance, while considering workload changes and server failures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If tenants are consolidated onto the same machine to improve utilization, then operational cost decreases, but service-level objectives for tenants may not be met
Solution Approach 1:
The patent segments the placement problem into two distinct phases: a configuration phase that determines optimal tenant-to-server assignments using predictive performance models, and a runtime phase that executes migrations based on those assignments. This segmentation allows SLOs to be guaranteed during configuration while enabling dynamic adaptation at runtime, resolving the contradiction between consolidation benefits and service level guarantees.
Solution Approach 2:
The patent performs preliminary actions by pre-calculating optimal tenant placements using predictive models during a configuration phase before actual migrations occur. This preliminary optimization ensures that when migrations happen, they are based on proven performance predictions, thereby maintaining SLOs while achieving cost savings through consolidation.
2Ease of manufacture
If static placement algorithms are used to assign tenants, then implementation is simple, but tenant migration costs are not considered
Solution Approach 1:
The patent incorporates feedback mechanisms by continuously monitoring actual tenant performance and comparing it against predictive model expectations. This feedback loop identifies when migrations are needed and guides the placement optimization process, ensuring that migration decisions are data-driven rather than arbitrary, thus reducing unnecessary migration costs while maintaining implementation feasibility.
Solution Approach 2:
The patent changes the parameter of tenant placement from static to dynamic by introducing predictive performance models that estimate how tenants will perform on different servers. This parameter change enables the system to consider migration costs by evaluating multiple placement scenarios and selecting those that minimize overall operational expenditure while maintaining service levels.
3Speed
If in-memory databases are used to speed up data access, then query performance improves, but memory capacity becomes the limiting factor
Solution Approach 1:
The patent applies multi-functionality by designing a predictive performance model that simultaneously evaluates multiple criteria: query performance (speed), memory capacity utilization, and operational cost. This universal evaluation framework allows the system to optimize for speed while consciously managing memory capacity constraints, rather than treating them as separate conflicting objectives.
4Loss of energy
If the number of servers is reduced to minimize costs, then operational expenditure decreases, but flexibility to handle load changes is reduced
Solution Approach 1:
The patent introduces dynamics by implementing a two-phase approach where the configuration phase establishes optimal placements for cost efficiency, but the runtime phase enables dynamic migrations in response to changing workload conditions. This dynamic capability allows the system to maintain flexibility and adaptability even with a reduced number of servers, as tenants can be relocated to handle load changes without requiring a large static infrastructure.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Present application refers to a robust tenant placement and migration system, process and to a respective controller in database-as-a-service environments. Multiple tenants (T) are to be allocated on a set of in-memory database servers (S) such that workload changes depending on the tenants' behavior are considered, while maintaining strict service level objectives. The placement and migration is based on a migration command which is based on an interleaved placement by interleaving tenants (T) across machines.