Presentation on theme: "Forecasting Data Center Power Requirements: Tips from the Trenches Frank Bereznay Kaiser Permanente."— Presentation transcript:
Forecasting Data Center Power Requirements: Tips from the Trenches Frank Bereznay Kaiser Permanente
Agenda Overall Facility Utilization / Usage Profiles Measuring Power Usage at the Equipment Level Reporting and Forecasting Techniques Summary
Facility Overview Two Story Facility Built in the early 1990’s to accommodate mainframe class equipment. 2 UPS Systems Primary UPS (A/B) 2,400 KW Secondary UPS (C/D) 600 KW Added later 3,500 Items Connected to UPS Systems Equipment Mix Described as typical for today’s Data Center. Newer blade farms creating hot spots for A/C systems
Initial Measurements There were a lot of opinions about the overall utilization profile for the UPS systems, but nobody had hard empirical data. A low tech ‘clip board’ agent was installed and activated at the UPS panels at four hour intervals.
Measuring Power Usage at the Equipment Level Not as easy as it sounds. Engineers were reluctant to open up the back side of PDU cabinets to attach reading devices. Power cables coming out of the cabinet were not labeled. Similar situation at equipment end of power run.
Measuring Power Usage at the Equipment Level Looked for some equipment documentation on actual power usage. Generally speaking, the documentation only provided rating factors for the circuit breakers. Blade servers were the exception, they provided actual power calculators. Looked online (and the CMG proceedings) for articles providing this kind of information. Came up empty.
Measuring Power Usage at the Equipment Level Pursued two efforts in parallel Captured before and after power readings at the UPS level for any change that projected a 5KW or greater impact Worked with staff at a newer Data Center that had the capability to provide circuit level power usage profiles Developed spreadsheet of findings and created some initial forecasts
Measuring Power Usage at the Equipment Level Initial forecasts tended to underestimate the impact of power events. Part of the learning curve for this type of analysis. A ‘life cycle’ usage profile for equipment types began to emerge from the dual measurement efforts being conducted.
Reporting and Forecasting Techniques All data capture and analysis was done in EXCEL, nothing fancy. Experimented with a variety of reporting techniques before we came up with the final set of reports.
Summary Primary goal of the paper was to share some lessons learned. Power analytics are fairly straightforward. The data can be elusive and difficult to get. The paper contains a table of power factors developed over time.