Columbia
Architecture and Design
Lab
vbench: a Benchmark for Video Transcoding in the Cloud
Overview
vbench is the first benchmark for the emerging video-as-a-service workload and will be presented at ASPLOS 2018 [1].
Current estimates place
video at 71% of evening traffic in North America and project that this
number will grow to 80% by 2020. Each video that is streamed over the
internet has been transcoded at least once by a cloud video provider such
as YouTube or Facebook. This creates a great computational burden on the
infrastructure of these companies. Furthermore, the rate of growth of
uploaded video content has now surpassed the rate of growth of
computational power for standard CMP systems. New solutions are necessary
in this space.
vbench captures the nuances of this important workload including the
tradeoffs that cloud video providers have to navigate between Quality of
Experience (QoE), network cost, compute cost, and storage cost.
Download
vbench is based on open source software and freely redistributable video sequences.
We would like to keep track of who is using vbench so please complete the following form to download the benchmark.
If you have questions on the usage of the benchmark contact ude.aibmuloc.sc.stsil@hcnebv
Support
This material is based upon work supported by 2016 Google Faculty Research Award, NSF (CCF-1065338) and C-FAR, one of six centers of STARnet, a Semiconductor Research Corporation program sponsored by MARCO and DARPA. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of these sponsors.