Emerging high-performance computing (HPC) systems show a tendency towards heterogeneous nodes that are dense with accelerators such as GPUs. They offer higher computational power at lower energy and cost than homogeneous CPU-only nodes. While an accelerator-rich machine reduces the total number of compute nodes required to achieve a performance target, a single node becomes susceptible to accelerator failures as well as sharing intra-node resources with many accelerators. Such failures must be recovered by end-to-end resilience schemes such as checkpoint-restart. However, preserving a large amount of local state within accelerators for checkpointing incurs significant overhead. This trend reveals a new challenge for the resilience in accelerator-dense systems. We study its impact in multi-level checkpointing systems and with burst buffers. We quantify the system-level efficiency for resilience, sweeping the failure rate, system scale, and GPU density. Our multi-level checkpoint-restart model shows that the efficiency begins to drop at a 16:1 GPU-to-CPU ratio in a 3.6 EFLOP system and a ratio of 64:1 degrades overall system efficiency by 5%. Furthermore, we quantify the system-level impact of possible design considerations for the resilience in GPU-dense systems to mitigate this challenge.
Published manuscript249.26 KB