HotCarbon 2024
Conference took place this year for the third time on the July 9, 2024.
HotCarbon aims to bring together researchers and practitioners in computer and networked systems to engage in a lively discussion around sustainability throughout the entire computing lifecycle, focusing on both the operational and embodied impact of computer systems.
Program: https://hotcarbon.org/program
For me personally the following papers were most interesting…
A Call for Research on Storage Emissions
→pdf
→slides
→My Zotero Highlights
- SSD rack vs. HDD rack:
- ~4× the operational emissions per TB
- ~10× the embodied emissions per TB
Energy-Aware Process Scheduling in Linux
- CFS scheduling in Linux is time-based (unfair regarding energy consumption)
- Experiment with two processes that are scheduled almost identical CPU time:
- Process 1: memory-intensive (repeatedly reads and writes large byte-arrays into heap memory)
- Process 2: CPU-intensive (repeatedly computes SHA256 on long strings)
- Result: process 2 consumes about 50% more energy than process 1
- possible reasons: usage of different CPU instructions, memory consumption, and the cache hit rate
- Uses eBPF functions to efficiently measure per-process energy consumption at millisecond-scale granularity with low overhead (1.25 %)
A Sleep Study for ISP Networks: Evaluating Link Sleeping on Real World Data
- Networks are built with the worst-case scenario in mind → network utilization is under-utilized (~3 %)
- Low utilization of network components are still consuming a lot of energy (not power proportional)
- Proposal: Turn off unused links to save energy
- Around 1/3 of links can be turned off on average
Bridging the sustainability gap in serverless through observability and carbon-aware pricing
- Serverless applications can have a significant carbon footprint depending on configurations and underlying infrastructure
- None of the existing studies/projects is effective in capturing and modeling the carbon emissions of serverless functions due to not considering unique serverless characteristics such as event-driven invocation, pay-as-you-go billing model, short execution duration, ephemeral runtime sandbox, and opaque underlying infrastructure
- pay-as-you-go billing enables cost-effectiveness, but fails to reflect a function’s environmental impact in its cost
- Proposals:
- improved carbon estimation methodology for serverless settings
- new carbon-aware pricing model
- Collect resource allocation metrics with eBPF and OpenTelemetry
BoaviztAPI: A Bottom-Up Model to Assess the Environmental Impacts of Cloud Services
→pdf
→slides
→video
→My Zotero Highlights
- considered embodied carbon is very important
- modeling of a cloud instance:
- share of the technical and building environment impacts
- share of each of the servers’ components impact (CPU, RAM, disks, motherboard, PSU, assembly, and casing)
- methodology of Boavizta seems to be quite good