Azure ML Goes to Southeast Asia

This blog post is authored by Hai Ning, Principal Program Manager at Microsoft.
NOTE: Since the time this post went live, we have also deployed to Western Europe, on August 31st, 2015.

Today, we are happy to announce that we are taking the first step towards making Azure Machine Learning available in Microsoft Azure data centers outside of United States. Specifically, we are enabling Azure ML deployments in Azure data centers located in the Southeast Asia region (Singapore).

Prior to this announcement, all Azure ML assets were located in US South Central region Azure data centers. This creates challenges for customers who are under regulatory constraints which require all their data storage and computation resources located within certain geographical boundaries outside the US. In addition, network latency and bandwidth limitations are also potential obstacles our international customers face. That’s why we’ve been working hard to make Azure ML available in additional regions, and – starting today – customers will be able to deploy Azure ML solutions in Microsoft’s Southeast Asia data centers.

To get started, go to your Azure management portal and first create a Storage Account in Southeast Asia if you don’t have one already. Alternatively, you can get one automatically provisioned for you while you create your Southeast Asia Azure ML workspace. (I always like to create a Storage Account on my own – makes me I feel I am in control!)

Next, create a new Azure ML workspace. Please ensure you select “Southeast Asia” in the location dropdown. And supply the Southeast Asia Storage Account created in step one. Note the Storage Account must be co-located together with the workspace in the same region to guarantee minimal latency. This is why the Storage Account dropdown only contains Storage Accounts created within the specified region.

Once the workspace is created, you can click on the “Sign in to ML Studio” link or the “Open in Studio” command bar button to access it.

   

A better/faster way to navigate to your workspaces, is to directly go to the Azure ML Studio homepage (https://studio.azureml.net) and log in. You will see all your workspaces listed in the top left workspace selector control. You can now then easily switch region and workspaces under the selected region. 

Once you select the workspace you created in the Southeast Asia region, all experiments, datasets, and web service you create will be located in that region, and all computation happens in the computer clusters in the data centers of that region as well.

When you open an experiment from the Azure Machine Learning Gallery, you will now also need to select a Region before selecting a workspace under that region in order to make a copy of that experiment.

 

There a few known limitations that we are actively working on and they will be addressed in timely way:

  1. With this release, you can only copy experiments between workspaces that belong to the same region. In the future, we will enable copying experiments between workspaces across multiple regions.

  2. You can only list workspaces under one region at a time in the workspace selector. In the future, you will be able to see a full list of workspaces you have access to across all regions at the same time.

  3. If you use a Free workspace or a Guest Access workspace from the Studio homepage, these workspace will continue to be created and operated out of the US South Central region. In the future, you will be able to create Free and Guest Access workspaces in the region that is determined to be more optimal.

  4. If you deploy a web service from a predictive experiment, the web service endpoints can only live in the same region that the experiment is created in. In the future, you will have the flexibility of creating experiments in one region, and deploying generated web service endpoints into different regions.

Microsoft currently operates many Azure data centers across the US, Asia, Europe, Latin America and Australia. We are gradually but aggressively making Azure ML available in all these regions so that customers with data sovereignty requirements can deploy ML solutions with the assurance of compliance within the geographical boundaries of the jurisdictions in which they operate.

Soon after this Southeast rollout, our next target is Western Europe based on our customer demand. If you have suggestions for future expansion in your region, please do let us know – we are always listening!

Hai