A tool for quantifying the total CO2 cost of ownership of database servers.
A Case for Ecological Efficiency in Database Server Lifecycles
Users can configure CPU, DRAM, SSD, and HDD of an existing server to compare it with another setup and asses its CO₂ emissions relative to its own embodied carbon.
This section enables users to modify the type of workload, the percentage of server utilization, and the intensity of carbon of the grid based on the location of the server.
Workload Types:
- SPECrate – Measures multi-threaded performance, simulating environments such as databases and web servers
- SPECspeed – Evaluates single-threaded performance for general purpose tasks such as data compression and text processing. We use publicly available measurements for both, SPECrate and SPECspeed.
- Sorting – A common yet computationally challenging task that is difficult to fully parallelize. A vector of four billion random integer values (uint32_t, 16GB) is generated, then the time to sort the entire vector is measured.
- TPC-H – Assesses analytical database performance by running TPC-H workloads with a scale factor of 10 and 25 read-only query streams on the open-source in-memory database system Hyrise [5]. We collect measurements for Sorting and TPC-H experimentally in our lab
Other Settings:
- Server Utilization – Defined as the ratio of queries per second to the maximum possible queries per second. According to the findings of Barroso and Hölzle, who monitored thousands of Google servers over six months, servers typically operate at between 10% and 50% of their maximum theoretical capacity rather than being idle or running at peak levels.
- Grid Carbon Intensity (GCI) – The GCI also plays a crucial role in predicting the ecological impact of upgrading components. The carbon intensity of a country’s power grid measures the CO2 emissions per kilowatt-hour of electricity produced.
The break-even time is visualized on a line chart, allowing users to assess the accumulated CO2 emissions across different configurations.
Additional key data points, such as break-even time, grid carbon intensity, embodied carbon of new hardware, total carbon footprint until break-even, workload performance indicator, and breakdowns of the embodied and operational carbon footprint are provided to give further insights into each comparison. These data points along with the line chart are then dynamically updated to reflect any changes made to the parameters.
| File | Description |
|---|---|
| cpu2006-results-20240723-164205.csv | SPEC CPU 2006 |
| cpu2017-results-20240723-171407.csv | SPEC CPU 2017 |
| tpc_cpu.csv | TPC-H/C most used CPU + TDP |
| intel_cpus.csv | Intel CPU information crawled from https://ark.intel.com |