Evolving from First In, First Out
When IonQ launched our first commercial quantum computer on the cloud in 2020, our queuing system was based on a weighted First In, First Out (FIFO) logic. In the last four years, we have launched two more classes of commercial systems and scaled systems within those classes to help meet growing customer demand. As we continue to scale to meet that demand, we are moving to a more advanced queuing system that will result in more fair access across all of our customers. We feel this customer-centric approach is aligned with our vision for the future of our cloud services.
Fair Share Queuing Approach
Going forward, the IonQ Quantum Cloud queue system will be managed by a fair share scheduling algorithm. Fair share logic has a precedent in the classical compute industry, and we are applying the learnings from those algorithms to develop IonQ’s new access model. In principle, this new algorithm is designed to:
Better Serve Our Customers | With our legacy model, FIFO, workloads can be split across the queue. This can result in time between jobs being executed, increasing the chance that system characterization changes can impact algorithmic results. |
More Effectively Balance Load | In times of high traffic, the queue can lengthen. Our new approach will help make sure that everyone waiting is getting an appropriate (i.e. “fair”) amount of time on our QPUs. |
Improve Queue Efficiency | In periods of high demand, small workloads can take a long time to run. With fair share queueing small, queue-based workloads can be mixed with other types of access to improve throughput efficiency. |
Enable New Scheduling Features | Our new queuing approach provides better support for our customers’ increasing demand for hybrid workloads, which require many positions in the queue. |
Reservations and Queue Access
Reservations are a way to obtain exclusive access to a QPU for a set period of time. This feature has been a critical component of how users run large tranches of jobs or hybrid workflows that require an iterative back-and-forth between the QPU and a CPU.
Reservations will continue to be available, and for the largest workloads this will continue to be a critical component. However, our new queueing algorithm allows us to start experimenting with more nuanced ways to allow users high-priority access for smaller workloads.
Keep an eye out for future updates and more information in the coming months from IonQ about how we’ll be using these new features to support smaller hybrid workloads across a variety of workflow types.
Determining Fair Share Queue Access
The new queuing system separates our customers into groups based on the capacity they have allocated on our systems and then provides each group a share of each cycle.
Cycling through these groups means that all our customers will get a chance to use our systems, and our customers’ jobs will never “stall” or get “stuck” in the queue, since all groups have a chance to get time on the QPU each cycle.
This new system also gives us greater control over our queue and provides new ways to manage the jobs within it. We’re currently working on ways to expand our options for managing the priority of your jobs in our system, and a special feature that will allow hybrid workloads to ensure that subsequent jobs will have access to the QPU after each classical portion completes.
Jobs submitted to the queue will run in accordance with the new fair share algorithm, which prioritizes jobs automatically based on predetermined logic.
Determining Allocations
Allocations are determined directly by the terms of our customers’ contracts. For example, a customer committed to “5% of an Aria-class system” will automatically have 5% of each loop dedicated to the jobs they have in the queue. Similarly, a customer buying 100 hours, starts with an allocation equal to 100 hours.
Learning More
This information is also available in our documentation for future reference, and if you have any additional questions, please contact your account manager or reach out to [email protected]