You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
13:00 Cluster at UZH (1 slide with some details about the cluster)
use these slides from morning part: efficientClusterComputation/docs/html/slides/content/hpcAtUZH/index.html
13:10 We login and have a look around (1 slide with helpful commands)
- 1 tab with slurm command
- 1 tab with unix commands
- df -hT
- ...
- see morning part: efficientClusterComputation/docs/html/slides/content/specialTopics/clusterCommands.html
13:30 Latency and Throughput (1 slide with the inter device latencies)
see morning part: efficientClusterComputation/docs/html/slides/content/recapHPC/nodeArchitecture.html
13:40 Profiling - we can make a slide with tabs:
- slurm tools for profiling
- unix tools fio for i/o (add a warning about throughput testing)
- network throughput wrk
- load average (btop or cat /proc/loadavg)
- try them out on your laptop (or VM)
- we try them on the VM in the cloud -> block (fio) vs object-storage (wrk)
- we try them on the cluster (maybe just 1 tries) -> shared file system (fio) vs object-storage (wrk)
14:10 Goal: Run the hello world on the cluster (1 slide that guides to "hello world exo", add the exercise there)
- track the input data with git lfs (1 tab)
- get the container from the object storage
- write the .sh submission script -> provide one with gaps to be filled in
- add data (results) upload to object storage
15:00 helloWorld exercise as embarrassingly parallel task (1 slide that marks this as homework and refers to the exercise repo. Mention on the slide that we should be added to the repo and see that you at least tried:-))
- We want to create 3 files that greet the participants for 3 different courses
- adapt the config to contain the 3 courses -> "checkout the branch that contains this already"
- introduce an env variable to select which course to use -> "checkout the branch that contains this already"
- adapt the .sh bash file to use multiple nodes and set the environment variable appropriately for each node
- 1 tab with slurm command
- 1 tab with unix commands
-
df -hT- ...
- see morning part: efficientClusterComputation/docs/html/slides/content/specialTopics/clusterCommands.html
- slurm tools for profiling
- unix tools
fiofor i/o (add a warning about throughput testing)- network throughput
wrk- load average (btop or
cat /proc/loadavg)- try them out on your laptop (or VM)
- we try them on the VM in the cloud -> block (fio) vs object-storage (wrk)
- we try them on the cluster (maybe just 1 tries) -> shared file system (fio) vs object-storage (wrk)
- track the input data with git lfs (1 tab)
- get the container from the object storage
- write the .sh submission script -> provide one with gaps to be filled in
- add data (results) upload to object storage
- We want to create 3 files that greet the participants for 3 different courses
- adapt the config to contain the 3 courses -> "checkout the branch that contains this already"
- introduce an env variable to select which course to use -> "checkout the branch that contains this already"
- adapt the .sh bash file to use multiple nodes and set the environment variable appropriately for each node