Georgia Tech inventors have designed a method to reduce the length of waiting queues by controlling the priorities of the multiple job types. The basic idea is giving high priority to the queue for light job types than the queue for heavy job types to avoid long waiting time for those quick jobs. The goal is to stabilize the fluctuation of response times seen in N-tier systems under a partial-saturation state. This priority control is only turned on when hardware resources are in a partial-saturation state; if the priority control was always on, there is a possibility of degrading the total performance of the n-tier system. The method autonomously and independently diagnoses the saturation of downstream tiers, meaning it does not require information from hardware that may already be partially saturated.
- Quick feedback on dynamic priority control of job execution
- Stabilize fluctuation of response times
- Develop applications
- Server software
The quick fluctuation of response time under the partial-saturation state of hardware is due to hardware resources reaching saturation, causing queues to form for incoming jobs. These queues have no priority regarding the various jobs, causing very short jobs to wait behind larger loads. Long wait lines take up software resources, which can bleed into upstream tiers, and produce additional waiting lines for software resources.