On Chip Controller(OCC) Overview Todd Rosedahl, Chief Engineer - PowerPoint PPT Presentation

On Chip Controller(OCC) Overview Todd Rosedahl, Chief Engineer IBM/POWER Firmware Development #OpenPOWERSummit Join the conversation at #OpenPOWERSummit 1

Agenda  Introduction/Motivation  System Stack/Ecosystem  Video  OCC Overview  Hardware Block Diagram  Firmware Details  Sample Results – Graphs  Future Enhancements  References Join the conversation at #OpenPOWERSummit 2

Motivation  US datacenter energy consumption – 91Billion KWH  34 500MW power plants  State of Minnesota – 68Billion KWH  Power Usage Effectiveness Join the conversation at #OpenPOWERSummit 3

Ecosystem Enablement July 2014 Power8 open source firmware stack Power Open Source Software Stack Components contributed thru GitHub Cloud Software Toolkits and resources for porting and optimizing, Existing Standard Operating growing repository on website Open Source Environment Software (System Mgmt) Communities Firmware Technical Training – OpenPower Pavilion Operating Thursday 9-11am Architectural Overview System / KVM Thursday 12-2pm Hands-on Training New OSS OpenPOWER Firmware Firmware Community System OpenPOWER Hardware Technology KVM/Linux OCC OPAL BMC Host Boot SBE Management Console ETH P8 HW Join the conversation at #OpenPOWERSummit 4

Your logo here Join the conversation at #OpenPOWERSummit 5

OCC Overview What is OCC?  Hardware/Firmware that controls power, performance & thermal  405 processor with 512k dedicated RAM  General Purpose Engines (GPE) to offload the 405 Fan Control Actuation What does OCC do? BMC GPE  Reads/controls system power  Memory Reads/controls chip temps Register Communicates Reads/Writes • Loads Enables efficient fan control Measures Actuates • Processor Provides OT protection Runs on Power Reads C OCC  405 Power Capping C Measurement  Fault Tolerance Measures/ Writes Actuates  Energy saving VRMs  Communicates Performance boost Communicates to Processor other OCC via Loaded and initialized by C 405 C Opal HostBoot Sys Mem Physical paths not shown Join the conversation at #OpenPOWERSummit 6

OCC Overview Fan Control Actuation BMC GPE Memory Register Communicates Reads/Writes Loads Measures Actuates Processor Runs on Reads C Power Measurement OCC 405 C Measures/Actuates Writes VRMs Communicates to other OCC Communicates Processor Loaded and initialized by C 405 C Opal HostBoot System Memory Physical paths not shown Join the conversation at #OpenPOWERSummit 7

Hardware Block Diagram Main Memory Uses  Initial Load  OPAL communication  OCC communication Main Memory Measurement  Temperature Sensing  Utilization Measurement  Bandwidth control Processor  Temperature Sensing  Core Frequency Control - PSTATES  Chip Voltage Control – PSTATES  Utilization Measurement BMC  Report power/temperature  Provide Power Cap  DCMI compliance Join the conversation at #OpenPOWERSummit 8

OCC Internal Details Join the conversation at #OpenPOWERSummit 9

PSTATE Details Linux Guest Linux Guest (LPAR) (LPAR) Utilization and SLOF SLOF governor policy are inputs to cpufreq in Linux Hypervisor libvirt Qemu Qemu Hypervisor Host Power Mode cpuidle cpufreq Policies Linux Host Kernel Cpufreq – exploits SPR Platform Drivers based P-state change PORE OPAL OCC mem OCC POWER8 BMC Host Boot System Join the conversation at #OpenPOWERSummit 10

User Set Power Limit Join the conversation at #OpenPOWERSummit 11

Processor Over Temperature Join the conversation at #OpenPOWERSummit 12

Potential Future Enhancements  Power Shifting Improvements • Memory Bandwidth % vs Processor Frequency • Power Shift between other devices  Modify Controls • Run processors at higher speeds • Run processor/memory at higher temperatures  Adjust loop times • Increase responsiveness to workload changes Join the conversation at #OpenPOWERSummit 13

References Open Power Blog link:  http://openpowerfoundation.org/press-releases/occ-firmware- code-is-now-open-source/ GitHub pages:  OpenPOWER project: https://github.com/open-power  OCC: https://github.com/open-power/occ  Building OpenPOWER: https://github.com/open-power/op-build Technical Firmware Training – OpenPOWER Pavilion:  Thursday 9-11am Architectural Overview  Thursday 12-2pm Hands-on Training – Build/Modify Reference Platform (Palmetto) – TYAN GN70-BP010 Join the conversation at #OpenPOWERSummit 14