Closed ShuyangCao closed 2 years ago
Hi @ShuyangCao. Thanks for your interest. The feature you asked is already available in this branch (via ts -i
. It would take a bit more time to merge into master but it is stable enough to be used now.
Thanks. Can I reinstall ts using this branch without killing the server?
Unfortunately it's impossible 🙁
Thanks @justanhduc. It works great! Just wondering, does this branch support 0 wait time? As the scheduler has recorded the GPU ids, I think it probably can avoid assign jobs to the same GPU, right?
Hi @ShuyangCao. Yes it does! Now the server will assign GPUs to client instead of letting client choose as before, so queued jobs are executed immediately if there are available slots and GPUs.
Thanks for developing the tool! Right now
ts -i
only returns the number of GPUs allocated andts -p
returns the pid of the main process. To know which GPUs the process and its child processes actually uses, I am usingpgrep
and crossing ref withnvidia-smi
. Is there an easy way to do so?