· boston university slideshow title goes here double dynamic provisioning (ddp) at the...
TRANSCRIPT
<Insert Picture Here>
© 2010 Oracle Corporation – Proprietary and Confidential
Kalyan Vaidyanathan and Kenny GrossOracle Physical Sciences Research Center
Intelligent Power Monitoring and Management for Enterprise Servers
© 2010 Oracle Corporation – Proprietary and Confidential 2
BACKGROUND: CONTINUOUS SYSTEM TELEMETRY HARNESS (CSTH)
• Original Oracle Telemetry Invention - US Pat. 7,020,802 (Mar 2006)
• “Real-time Telemetry System for Enterprise Computing Servers For Enhanced Availability, QOS, and Security”
• CSTH has spawned a portfolio of 80+ Oracle patents
• Proactive fault monitoring (Electronic Prognostics)
• Enhanced energy efficiency (Intelligent Power Monitoring, Intelligent Fan Control, Intelligent Energy-Aware Workload and Cooling Provisioning)
• In-situ vibrational integrity characterization for servers and storage
• Reduction/Avoidance of Worldwide FCOs (product recalls)
© 2010 Oracle Corporation – Proprietary and Confidential 3
“Soft” VariablesSystem Performance Variables [Sources: kstat, application s/w]
“Canary” VariablesDistributed Synthetic Transaction Generators (user transaction latencies, monitored 24x7)
“Black Box” RecorderCircular File Structure.Retains high sampling rate signals 72 hrs; lower sampling rate signals for life of server
Proactive: Predictive Failure AnnunciationReactive: Faster, more accurate Root Cause AnalysisSelf Healing: Software performance issues (e.g. resource
contention issues, memory leaks)IPM: Inferential Power monitoringIFC: Intelligent Fan Control
Physical VariablesDistributed internal temperatures, currents, voltages, fan speeds, vibrations
Advanced patternrecognition }
Continuous System Telemetry Harness
© 2010 Oracle Corporation – Proprietary and Confidential 4
Examples of time series telemetry data
Intelligent Power Monitoring (IPM)
Current
Voltage
Temperature
Physical Parameters
OS ParametersCPU, Memory, I/O traffic, Disk Utilization, etc.
Response Time, Trans-action Latency, etc.
Application Parameters
IPM Data Collector
Accuracy in IPM estimatesReads PSU current and voltage sensors to compute power
Uses NLNP estimation with various telemetry variables
Substantially better power estimation (factor of 7 accuracy improvement) by use of IPM
Why Dynamic Power Monitoring for Servers
Realistic Power Draw
Continuous Power Draw
Readings
Ease of Installation
Server Level Power Information
Correlation of Power with System &
Environmentals
Nameplate Rating
Online Power Calculators
Handheld Power Meters
Networked External Power MetersRack-LevelPower Monitoring (Metered PDUs)Server-level Dynamic Power Monitoring
N/A
N/A
Power Monitoring Methods
Advantages
Actionable Power Savings from IPM ServiceInsight from Correlated Data
Boston University Slideshow Title Goes Here
Thermal Profiling: Rack of Servers
Boston University Slideshow Title Goes Here
Data Center Thermal Flux MappingContinuous real time dynamic thermal flux and power fluxinside the server assets.
*Oracle Patent: “Datacenter Spatial Thermal Flux Mapping Via Telemetry,” U.S. Patent 7,549,070
Boston University Slideshow Title Goes Here
Double Dynamic Provisioning (DDP)At the rack/datacenter level, DDP provisions load preferentially tothe cool spots via (thermal-aware virtualization workload mobility) and cooling preferentially to the hot spots.
Optimal energy efficiency while minimizing spatial and temporal thermal gradients (resulting in improved long-term reliability of data center assets).
"Optimized Workload Scheduling For Improved Energy Utilization and Reliability for Multicore Chip Servers," U.S. Patent 7,716,006"Double Dynamic Provisioning Method and Apparatus for Optimal Datacenter RAS and Energy Utilization," (Oracle patent pending).Telemetry-Enabled Energy-Aware and Temperature-Aware Scheduling for Optimal Server Energy Utilization and Reliability," (Oracle patent pending).
Boston University Slideshow Title Goes Here
Oracle Intelligent Fan Control (IFC)
Continuously “seeks and settles” at optimal energy minimum, balancing cubic fan power vs exponential leakage power
Attains optimal server energy without penalizing server performance (as does CPU frequency scaling, clock gating, and DIMM memory throttling).
*Oracle Patent:"Optimal Fan Controller for Eco-Efficient Computer Servers," U.S. Patent 8,108,697
13
Internal component temperatures with no IFC
IFC Engaged in Firmware
IFC: Internal temperatures rise to their “comfort zone” (defined by reliability specs)
Continuous internal system telemetry enables Intelligent Fan Control to collapse excessivethermal headroom margins the industry has heretofore been blind to, mitigating wasted energyfrom fan motors, lowering vibrations and acoustics.
Time(min)
14© Oracle and/or its affiliates – Proprietary and Confidential
© 2010 Oracle Corporation – Proprietary and Confidential 15
Appendix