One of the (many) great aspects of my current role working in the Innovations area of a UK Bank is a relentless introduction to new features in Microsoft Azure. At my stage with Azure in practice and exams it is usually a new feature or behaviour that has dropped as part of a generation 2 (E.g. Storage vs Data Lake) or evolution of features or more subtly a change to the defaults of a combination (e.g. Automation and Desired State Configuration Extension). Then there are the “never heard of it” moments when a term gets mentioned and I rattle straight to a search engine.
One of these a few months back was Azure Cyclecloud, one of our projects involved input from Microsoft and their HPC specialist proposed it as a key component of the platform being evaluated. In our case it is acting as an orchestrator / scheduler and keeping tabs on a handful of low priority virtual machine scale sets.
I’ve not had any direct exposure to HPC beyond awareness due to Microsoft architectural exams I’ve done in the past for on-premises Windows, and latterly Azure cloud. The good news is within parameters that the news is good and Azure CycleCloud appears straightforward and being predominantly IaaS based is fairly easy to secure within our patterns. My thoughts so far are:
- The web admin interface is fairly sensitive to environment – I’ve lost about a day to Internet Explorer (doesn’t work) and the reverse proxy on our firewall appliances mangling page scripts.
- The manual install is straightforward and reliable in my limited experience – we have a vnet model that it sits in quite nicely and the documentation is good on required ports and cluster communications.
- Azure Cyclecloud being HPC and batch etc comes from open source land, so lots of command line and Linux – quite ironic that my career includes so many loops (my first job at an accountants in the 1980s included being the guy who wrote sql reports using vi on a unix practice management system).
- Following on from the previous point, Azure Cyclecloud integrates with Active Directory and therefore has it’s own RBAC model – very important to understand if you are trying to secure it.
- I have a few concerns about the quickstart deploy, mainly due to the public ip address bound to a server but that probably reflects our use cases and my background. (Googling “cyclecloud initial setup” reinforces this concern as a number of servers in initial setup pop up).
- The cloud account relies on a fairly big service principal so it’s important to keep on top of that bearing in mind the last two points.
- About 50% of the time I get the name wrong and call it Azure CloudCycle. This hit rate is slowly improving.