Lab 11d: Storage: GPUDirect Storage on lab-gpu01 (Lab 11 Part D)
60 minutes suggested, up to 120 in one session. Price: 5 credits per minute.
Hosted, this lab covers Parts D. The other Parts need your own machine (Home Lab Kit) and are shown collapsed in the lab.
The GPU tier is not available on this site yet.
What you will be able to do
- Build a local tier and an NFS tier on one VM, and prove which tier a path is on with
findmntandnfsstat -m. - Write
fiojob files that measure sequential bandwidth, small random reads and file creation, and keep them as a baseline. - Show how the page cache flatters a benchmark, and write jobs that measure storage instead of memory.
- Find the NFS client options that slow writes down, from the client's own statistics, and fix them so the fix survives a reboot.
- Install GPUDirect Storage (GDS), read
gdscheck.py -p, and prove withgdsiowhether the direct path or compatibility mode is in use. - Explain how
/etc/cufile.jsonandCUFILE_ENV_PATH_JSONwork together, and find a setting that comes from the environment rather than the file.