User Tools

Site Tools


cluster:198

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Next revision
Previous revision
cluster:198 [2020/12/02 09:21]
hmeij07 created
cluster:198 [2020/12/03 08:09] (current)
hmeij07 [GPU checkpoint/restart]
Line 4: Line 4:
 ===== GPU checkpoint/​restart ===== ===== GPU checkpoint/​restart =====
  
-Why I thought this was an easy problem to solve I do not know. CPU checkpoint/​restart has come a long way with [[cluster:​190|DMTCP]] for serial and parallel jobs (including multi-host). ​+Why I thought this was an easy problem to solve I do not know. CPU checkpoint/​restart has come a long way with [[cluster:​190|DMTCP]] for serial and parallel jobs (including multi-host). But the CPU/GPU environment adds much complexity.
  
-A good overview of the history of GPU checkpoint/​restart efforts can be found at this presentation+A good **overview** of the history of GPU checkpoint/​restart efforts can be found at this presentation
  
   * [[https://​on-demand.gputechconf.com/​gtc/​2016/​presentation/​s6429-akira-nukada-transparen-checkpoint-restart-technology-cuda-applications.pdf|Transparent Checkpoint and Restart Technology for CUDA® applications]]   * [[https://​on-demand.gputechconf.com/​gtc/​2016/​presentation/​s6429-akira-nukada-transparen-checkpoint-restart-technology-cuda-applications.pdf|Transparent Checkpoint and Restart Technology for CUDA® applications]]
 +
 +An **excellent** in depth explanation can be found in this article
 +
 +  * [[https://​arxiv.org/​pdf/​2008.10596.pdf|CRAC:​ Checkpoint-Restart Architecture for CUDA with Streamsand UVM]]
 +
 +  * git site (notice __early development__):​ https://​github.com/​DMTCP-CRAC/​CRAC-early-development
 +
 +Some quotes from the article ...  //CRAC provides the ability to save and restore the state of CUDA by first using CUDA-specific save/​restore operations, and then delegating to a traditional checkpoint-restart package.... In the end, the support of DMTCP for process virtualization and plugins makes it easier to add modular support for CUDA without having to excessively understand details of the internals of the host checkpointing package.// ​ This plugin will likely only work with recent gpu models.
 +
 +
 +  * Slurm DMTCP integration with plugin
 +    * https://​slurm.schedmd.com/​SLUG16/​ciemat-cr.pdf
 +  ​
  
 \\ \\
cluster/198.1606918863.txt.gz · Last modified: 2020/12/02 09:21 by hmeij07