1 / 11

Thermal Management in Datacenters

Thermal Management in Datacenters. Ayan Banerjee. Thermal Management using task placement. Tasks: Requires a certain number of servers (cores) for a specified amount of time. Each task has certain power consumption on each server of a particular node.

Download Presentation

Thermal Management in Datacenters

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Thermal Management in Datacenters Ayan Banerjee

  2. Thermal Management using task placement • Tasks: Requires a certain number of servers (cores) for a specified amount of time. • Each task has certain power consumption on each server of a particular node. • Assign cores to the tasks based on certain objective. • Remember task placement and not task scheduling

  3. Goals of this project • Build CFD model of a real Datacenter • Perform thermal profiling • Test the performance of different task placement algorithms on this model • Simulate Cross Interference minimization algorithm for multiple task scenario arriving at different time intervals • Test the optimality of the solutions for different objectives

  4. CFD Models • Flovent 6.1

  5. Difficulties faced with Flovent • Simplified Heterogeneous datacenter takes 18 hrs to run the base case • Cannot set parameters dynamically • According to findings HVAC outlet temperature varies with its inlet temperature • Cannot simulate that in Flovent so there will be difficulties in simulating algorithms that try to reduce total energy consumption

  6. Observations • Energy aware task placement algorithms must take into account the behavior of the Cooling System • Basis: AC works harder when the Hot isle temperature in the datacenter increases

  7. Observations • For the objective of minimizing total energy • We have to consider the working of the AC in the objective function • We have to consider a heterogeneous Datacenter • We have to design algorithms that will allow different jobs to work on the same server

  8. Experiment • Took the simplified CFD model of the Datacenter • There were 50 chassis. The design was for homogeneous environment. • Built a heterogeneous environment with 20 chassis equipped with dual core processor and the rest 30 chassis with quad core. • Total number of cores = 1300. • Dual core – idle = 1728 W, Busy = 3260 W Quad core – idle = 2420 W, Busy = 6020 W • Two applications T1 and T2 • T1 required 288 servers for a time period of 3 units starting from unit 0 to unit 3 • T2 required 672 cores for a time period of 3 units starting form unit 1 to unit 4 • Find the solution for the cross interference minimization algorithm for the objective of minimizing Maximum Temperature and minimizing total energy.

  9. Task Placement Minimizing Maximum Temperature Maximum Inlet Temperature = 24.6016 degrees Total Power = 162525 W

  10. Task Placement Minimizing Total Energy Task 2 Task 1 Total Energy Consumption = 576592 J MaxTin = 26.3093 C 29.0858 C 27.7871 C

  11. Goals Revisited • CFD Model of Datacenter not yet ready • We have information on Saguaro Racks but little information on other racks • Certain physical parameters need to be recorded • Power Profiling not done as a result of incomplete CFD Model • Simulation Environment for multiple task arriving at different times ready • Optimality of the cross interference minimization algorithm tested • Apart form the cited goals a lot of observations useful for future work are made

More Related