By: Simon Farnsworth (simon.delete@this.farnz.org.uk), June 19, 2019 3:23 am
Room: Moderated Discussions
Reading Papers (clzero.delete@this.avx.com) on June 18, 2019 5:45 pm wrote:
> New paper from Facebook analyzing their workloads: https://research.fb.com/wp-content/uploads/2019/05/SoftSKU-Optimizing-Server-Architectures-for-Microservice-Diversity-@Scale.pdf?
>
The link to the PDF directly (above) was broken for me; https://research.fb.com/publications/softsku-optimizing-server-architectures-for-microservice-diversity-scale/ worked, and links to the PDF.
> Shows:
> - Instruction type breakdown (high FP content may surprise some)
> - Bottleneck breakdown
> - IPC per workload
> - TLB and cache miss rates
> - Bandwidth vs. latency per workload
> - Optimization experiments showing sensitivity to things like core count, frequency, cache sizes, etc.
>
> Nice set of data and compares/contrasts to Spec and Google's published data.
>
> New paper from Facebook analyzing their workloads: https://research.fb.com/wp-content/uploads/2019/05/SoftSKU-Optimizing-Server-Architectures-for-Microservice-Diversity-@Scale.pdf?
>
The link to the PDF directly (above) was broken for me; https://research.fb.com/publications/softsku-optimizing-server-architectures-for-microservice-diversity-scale/ worked, and links to the PDF.
> Shows:
> - Instruction type breakdown (high FP content may surprise some)
> - Bottleneck breakdown
> - IPC per workload
> - TLB and cache miss rates
> - Bandwidth vs. latency per workload
> - Optimization experiments showing sensitivity to things like core count, frequency, cache sizes, etc.
>
> Nice set of data and compares/contrasts to Spec and Google's published data.
>
