Utilization

Published on Mar 18, 2025, filed under management (feed). (Share this on Mastodon or Bluesky?)

Queueing theory is interesting. We can probably all appreciate how it’s interesting from the practical view of waiting lines, and how these can be optimized. But queueing theory is also interesting to optimize one’s own work, going beyond merely adding “buffer.”

I’m still exploring the topic. However, there’s already something that I’ve singled out to use with my own and my teams’ work—to aim for 80% utilization.

Allow me to take the convenient route here, and share just what got me to this idea and number. Complementing articles I’ve been reading about the subject, it was a chat with ChatGPT. ^*

Here are three general rules and rules of thumb for optimizing work handling both for you individually (one server) and for your team (multiple servers):

For you individually (one server system):

1. Utilization rule (ρ < 80%):
   • Rule: Keep your utilization below 80% to avoid excessive delays and burnout.
   • Why: High utilization can lead to long queues and increased stress. Keeping it below 80% allows for flexibility and unexpected tasks.
2. […]

For your team (multiple servers system):

1. Load balancing (distribute evenly):
   • Rule: Distribute tasks evenly across team members to balance the load.
   • Why: Ensures no team member is overwhelmed, leading to better overall performance and reduced wait times.
2. Queue discipline (priority queuing):
   • Rule: Implement priority queuing for tasks, where critical tasks are handled first.
   • Why: Helps focus on high-impact work, ensuring important deadlines are met while managing less critical tasks later.
3. Pool utilization (ρ < 85% for n servers):
   • Rule: Keep the overall utilization of the team below 85%.
   • Why: For multi-server systems, slightly higher utilization is manageable, but staying below 85% helps prevent excessive queuing and ensures quality.

I find this plausible and useful. I’m testing this and may follow up here or in another entry; and I love to learn more about your lessons from queueing theory and how we can make (not over-engineered) use of it in our individual, as well as in our teams’ work. What works? What does not?

^* How cliché!—and yet you and I know that this is today’s information worker’s world, in which AI is being used as part of the learning mix. As long as AI users consider a variety of sources and keep thinking critically, I’m not sure we can (or want to) uphold a stigma around using AI.