Energy-Efficient Computing and Data Centers

Conclusion

After describing the major IT components in a data center and their power consumption, we presented how servers are cooled and the challenge of air-cooling dense servers with the trend of increasing power of processors and accelerators leading to higher heat capacity per rack. We presented various cooling solutions for the data center leveraging air cooling and water cooling and how waste heat can be reused to cool the IT devices in the data center. We showed, through the recent evolution of processors and accelerators, that higher performance is still delivered every generation, but at the expense of an increased power consumption, an increasing number of cores and the introduction of more specialized processors.

We analyzed the power, temperature and performance of a server when running various workloads. We showed that the two major application characteristics to influence its power consumption are CPI (cycle per instructions) and GBS (total memory bandwidth), while the processor frequency is the major system characteristic. We analyzed the impact of cooling on power, temperature and frequency, and showed water cooling allows processors to run at lower temperature due to the high heat capacity of water, enabling servers to either consume lower power or run at higher frequency, leading to potential higher performance.

As accurate measurements of the of the IT devices power and performance is critical for the data center, we presented the low-level sensors and APIs ...