Red Hat Bugzilla – Bug 463836
[LTC 6.0 FEAT] 201195:Scaled utilisation metrics for power-managed systems
Last modified: 2010-10-18 15:17:19 EDT
=Comment: #0================================================= Emily J. Ratliff <emilyr@us.ibm.com> - 2008-09-24 13:52 EDT 1. Feature Overview: Feature Id: [201195] a. Name of Feature: Scaled utilization metrics for power-managed systems b. Feature Description Provide correct utilization metrics as a function of total CPU capacity. For example utilization may be relative to highest nominal state or to current power cap. This may also impact a number of system utilities (eg: ps, top, others...). Currently CPU frequency scaling for power management may cause CPU utilization metrics to become incorrect. 2. Feature Details: Sponsor: LTC Architectures: x86 x86_64 ppc64 Arch Specificity: Both Affects Core Kernel: Yes Delivery Mechanism: Direct from community Category: Kernel Request Type: Kernel - Enhancement from IBM d. Upstream Acceptance: Accepted Sponsor Priority 2 f. Severity: Medium IBM Confidential: no Code Contribution: IBM code g. Component Version Target: 2.6.25 3. Business Case Sophisticated enterprise-level system and cluster management products need accurate accounting to correctly manage CPU resource utilization versus power utilization. Increasingly the enterprise market is interested in mitigating and managing power consumption in servers, where interest previously was mostly in a mobile market. An enterprise Linux distribution which enables sophisticated monitoring and control applications will have an advantage in this emerging market. 4. Primary contact at Red Hat: John Jarvis jjarvis@redhat.com 5. Primary contacts at Partner: Project Management Contact: Sarah Wright, sarahw@us.ibm.com, 503-578-5145 Technical contact(s): Dipankar Sarma, dipankar.sarma@in.ibm.com Gautham R Shenoy, gautshen@in.ibm.com IBM Manager: Jeffrey Heroux, heroux@us.ibm.com
What configuration options, if any, are needed for this? How is it exposed to userspace for apps to use?
(In reply to comment #4) > What configuration options, if any, are needed for this? How is it exposed to userspace for apps to use? CONFIG_VIRT_CPU_ACCOUNTING=y once the chargeback accounting patches show up in mainline. The taskstats netlink interface is how one gets advanced utime/stime data; see getdelays.c for a simple example of how it works. The utime/stime numbers will be scaled down (or up) based on the p-state when the processor is in C0.
Assigning this to Peter Bogdanovic at IBM.
Support for Power processors upstream in 2.6.31. Support of Intel and AMD processors unlikely in RHEL6.0 because the cpus don't support it properly.
Reassigning to Darrick, please answer Red Hat's question and move bug to Submitted state afterwards. Since in RH cvs tree, moving to Accepted state