https://cs.lozic.me/areas/performance/Recent content in Performance on CS TheoremsHugoen-usFri, 17 Dec 2027 12:00:00 +0100https://cs.lozic.me/posts/t109-work-stealing-and-fair-scheduling/Fri, 17 Dec 2027 12:00:00 +0100https://cs.lozic.me/posts/t109-work-stealing-and-fair-scheduling/<h2 id="symptom">Symptom</h2> <p>You parallelize a recursive algorithm. Quicksort, or a tree traversal, or a divide-and-conquer numeric kernel. You have 16 cores and a central task queue, and you measure a speedup of 3.</p>https://cs.lozic.me/posts/t009-balls-into-bins/Fri, 12 Feb 2027 12:00:00 +0100https://cs.lozic.me/posts/t009-balls-into-bins/<h2 id="symptom">Symptom</h2> <p>You have 100 backends and a load balancer hashing request IDs to pick one. Expected load per backend is exactly 1%, and you have checked the hash is good.</p>https://cs.lozic.me/posts/t082-gustafsons-law-and-the-usl/Fri, 25 Sep 2026 12:00:00 +0200https://cs.lozic.me/posts/t082-gustafsons-law-and-the-usl/<h2 id="symptom">Symptom</h2> <p>Two measurements that both look like they break the previous post.</p> <p>The first: your Spark job takes four hours on ten nodes. You move it to a hundred nodes and it still takes four hours — but it is now processing ten times the data. <a href="https://cs.lozic.me/posts/t081-amdahls-law/">T081</a> said the ceiling was 20×, and nobody hit a ceiling.</p>https://cs.lozic.me/posts/t081-amdahls-law/Fri, 18 Sep 2026 12:00:00 +0200https://cs.lozic.me/posts/t081-amdahls-law/<h2 id="symptom">Symptom</h2> <p>The profile was unambiguous: 94% of wall-clock time in one loop, and the loop&rsquo;s iterations are independent. You parallelise it across sixteen cores, the flame graph flattens exactly as predicted, and the end-to-end time goes from 40 seconds to 17.</p>