slowness of azure ml cloud compute

Question

slowness of azure ml cloud compute

Xinrui Li 6

I am doing a speed comparison test between my machine and azure ml cloud compute. My PC is 8 cores with 64 GB RAM. The compute instance I created on azure is "Standard_D14_v2 (16 cores, 112 GB RAM, 800 GB disk)". One test I did was to run a XGBoost model and it took about 1h15m locally. On the cloud, it took 1h45m. I thought I would have a better performance with the instance created on azure. Could someone explain to me why I saw the opposite? Thank you so much!

Math Law 0 Reputation points

2023-07-14T14:35:33.6666667+00:00

Same for me , even i increase core/ram sizes , Cloud still slower than local limited PC , any guides to follow ?

UPDATE !

It turns out that a different version of one ml algorithm is causing the slowness. (XGBoost version mismatch )

It is useful to check version of the library, if you are not installing/enforcing with REQUIREMENTS.txt.

1 answer

Your answer

Math Law 0 Reputation points

2023-07-14T14:35:33.6666667+00:00

Same for me , even i increase core/ram sizes , Cloud still slower than local limited PC , any guides to follow ?

UPDATE !

It turns out that a different version of one ml algorithm is causing the slowness. (XGBoost version mismatch )

It is useful to check version of the library, if you are not installing/enforcing with REQUIREMENTS.txt.

Answer 1

romungi-MSFT 48,911 Microsoft Employee Moderator

@Anonymous One factor that could effect while running your experiment on Azure ML is the setup time of the compute while running the experiment. Normally, compute is setup run time when the experiment run is initiated to avoid costs of running compute when not required. In this case the difference is still 30 minutes which is large and the setup usually takes few minutes depending on the availability of compute and region you are using.

Another factor could be the settings used in the experiment, since you are using the SDK or designer the default settings might be set for some configurations for optimization. If these settings are not changed then these might slow down the run time of the experiment, depending on your configuration these settings can be reviewed to check if there is a possibility to optimize it further.

If the above scenarios do not help we would recommend to raise a support issue to check what could be causing slowness in your workspace. Thanks.

Xinrui Li 6 Reputation points

2021-04-20T00:57:21.423+00:00

Hi romungi,

Thanks for your reply. I think the first factor doesn't apply to my case. I did another test which took 4 hours to run locally and 7 hours on the cloud. The setup time was about 5 minutes.

As for the second factor, would you be specific about which configurations I need to change? Is there any tutorial I could follow? Thank you!
romungi-MSFT 48,911 Reputation points Microsoft Employee Moderator

2021-04-20T07:14:14.63+00:00

@Anonymous Yes, you can checkout this sample notebook. It shows an example to change the default values and then run the config for the experiment.

Share via

slowness of azure ml cloud compute

1 answer

Your answer