Building the Right Environment to Support AI, Machine Learning and Deep Learning
A recorded chat by security architect for Microsoft Azure, Charlie Kaufman was published earlier this week by Microsoft. In this chat session, Kaufman described the broad concepts that enable security for Microsoft Azure customers, and he conceded at one point that Microsoft Azure security is "secure enough for some applications and not secure enough for others." Basically Microsoft Azure tenants (term used for Microsoft Azure customers) access virtual machines that tap into Microsoft Azure's pooled resources in the internet cloud environment. Access to these services are tied to each user's account which is established through a subscription portal. Tenants gain access to the service through a Windows Live ID. Kaufman said that the "cryptography behind Live ID is good."
Microsoft Azure has three basic components: storage, compute and SQL Azure (another form of storage). All these components run on separate hardware and communication is established via HTTP or SSL requests. Although all of the data on Microsoft Azure is stored in a single pool, access is only enabled via a secret key for each account, Kaufman explained. Microsoft Azure uses a different file system for its multi-tenant architecture. Existing applications need to be modified to use different types of storage, mainly blob (Binary Large Object) storage, Kaufman said. The C:, D: and E: drives that users see actually are virtual hard disks in the root operating system. Inputs and outputs go to the root OS and it makes sure that customers can only talk to their own disks. A network packet filter protects users from attacks from the outside, he added. Some attacks are possible with Microsoft Azure. The customer administration interface could be used to launch attacks. However, Microsoft typically keeps watch by checking for any malformed requests.
Microsoft provided more information about security for Microsoft Azure.