TL;DR: Agent Manufacturing Systems do not need extensive server infrastructure to operate.
Over the past year, we have been running our Brokering Additive Manufacturing living lab with intermittent jobs being submitted by academics and students alike. The service has 10-15 machines connected at any one time. The system has been reliable (for the most part, it is an alpha research codebase after all, 😄) and robust with machines able to connect and disconnect without disrupting the system. The broker service that manages the communications between agents is currently running on an AWS EC2 t2.micro instance and has met our needs.
However, the long-term vision is of a network featuring hundreds of jobs and machines bidding and brokering work. So, a question that our project partners and collaborators ask is “how does the system scale?” and what level of compute resource is required. The challenge is that we cannot to examine this with out living lab. We simply do not have enough printers and jobs.
To answer this question, we enlisted the help of the Engineering Compute Services (ECS) team at the Centre for Modelling and Simulation (CFMS). CFMS have a private cloud instance as well as the experience and knowledge to perform scale up studies of engineering software. Using their platform, we created services containing a range of broker instances that had received connections from virtual machines and jobs. In effect, we made a Digital Twin of the system!
The results showed that the servers scaled well in terms of compute resource enabling a modest cloud instance to be used to for hundreds of machines and jobs. The challenge comes in the bandwidth in handling all the communications and where there may be peaks if a number of contracts are made at the same time resulting in surge of gcode being transmitted.
The findings have been crucial to our work and we’re continuing to explore how we can tailor the communication protocol to reduce bandwidth yet not impact the brokering and bidding for work. We are currently running experiments with the broker filtering communications based on metadata provided by the machines and jobs about themselves to reduce unnecessary communications.
We had a great time working with CFMS and can’t wait to do more studies like these in the future.