symbioticlab / oort Goto Github PK

Hello, I'm reproducing your code, what is "manager_port" in the parameter setting

Hi,

I notice the fedscale provides the data for oort. And the fedscale on GitHub has been updated. For example, the device info is splited into client_behave_trace and client_device_capacity while oort still uses client_profile.pkl. Besides, there are also some differences between the configurations for tasks.

To make the oort correct, I guess I have to use the old version of fedscale, right? Say, commit before May 24th. Or you'll update oort later?

Thanks.

#!/bin/bash

gdown -O ./speech_samples_f16.pkl https://drive.google.com/uc?id=1sAnJ-4zySG65JtT47XOXzBHc4zHhgz9C
gdown -O ./openimg_distr.pkl https://drive.google.com/uc?id=1WBQBra_wv2sLU18l4uiq3JgMRcudyfG0
gdown -O ./client_profile.pkl https://drive.google.com/uc?id=1F8qLzUdINPSH_pSsbcr81Ry7-sBY1DVB

These links show 404 error, please confirm if the files and access still intact.

Hi,

It said we can reproduce figs 11 and 12 in the readme. But I only get the plot_perf.py to reproduce the line figures.

Could you please provide the plot.py for the bar figures for fair comparison?

Thank you so much!

Best

Hello,

There is an issue with the execution of with the oort sampler which causes the program to stop.

Experiment: running google speech benchmark with the same configurations as set by default in the conf.yml in the repo.

Error: probabilities do not sum to 1 thrown by resampleClients function of the parameter server:

sampledClientsRealTemp = sorted(clientSampler.resampleClients(numToSample, cur_time=epoch_count))

Log output when running with 100 clients (stops every time at Epoch 24):

Log output when running with 10 clients (stops at Epoch 329):