Optimizing System Performance and Workflow: From Technical Metrics to DevOps Process Improvement

Many technical professionals are promoted to management positions without formal management training, leading them to optimize work using personal contribution methods rather than systematic, holistic approaches.

An example is presented where an image‑recognition program running on a 10‑server cluster cannot meet a daily target of processing one million images. Initial calculations show that each server can handle only 96,000 images per day, suggesting the need for about 11 servers.

Rather than immediately adding hardware, the article stresses the importance of measurement and analysis. By examining the program’s architecture—recognition and comparison functions executed serially—the resource utilization is identified as inefficient.

The proposed solution splits the original program into two separate services (recognition and comparison) communicating via a message queue, allowing each to run on its own server and fully utilize resources. Re‑calculations show that the recognition service alone would need about 6 servers and the comparison service about 5, still totaling 11, but the new design saves GPU cards and enables higher concurrency for the comparison function.

The discussion then broadens to enterprise workflow, describing a typical defect‑resolution scenario involving developers, testers, and operations, where lack of clear versioning, deployment, and environment management causes delays and wasted effort.

Key problems identified include unclear code baselines, poor release documentation, ambiguous versioning, slow infrastructure provisioning, manual deployments, and inadequate environment tracking.

The article argues that, similar to system performance tuning, organizations must first map out their entire workflow, measure each step, and apply visual analytics (e.g., Kanban, burn‑down charts) to locate bottlenecks. Applying lean and DevOps principles can dramatically improve efficiency.