The source code of the paper"Laius: Towards Latency Awareness and Improved Utilization of Spatial Multitasking Accelerators in Datacenters" in ICS 2019.
The link of the paper ：https://dl.acm.org/doi/10.1145/3330345.3330351
The Homepage of Wei Zhang (First author): https://olivia-zhang.github.io/
Laius, a runtime system that carefully allocates the computation resource to co-located applications for maximizing the throughput of batch applications while guaranteeing the required QoS of user-facing services.
Prerequisites: CUDA, CUDNN, Caffe1.0, Tonic suite, Rodinia.
/build: cmake .. make j8
/mps: ./cudnn_server.sh /schedule.sh