Closed mental2008 closed 1 year ago
Due to the diurnal pattern of application workloads, it's intuitive to co-locate the applications on the accelerators (e.g., GPUs, ASICs, and FPGAs) to improve the utilization and efficiency. But the current work doesn't address the problem of interference when co-locating applications on the accelerators. The problem is fundamentally different than contention on multi-core CPUs.
Though the research problem is interesting, some concepts are still hard to follow.
The design of Baymax is as follows:
Comparison between Baymax and prior work:
We also would like to acknowledge Xayden, born 5 weeks before the deadline, for being an awesome and easy baby.
Glad to see this ;-) Just curious whose baby this is?
Presented in ASPLOS '16. [ Paper ]
Authors: Quan Chen, Hailong Yang, Jason Mars, Lingjia Tang University of Michigan, Shanghai Jiao Tong University, Beihang University