============================================================ INTERPRETATION SNAPSHOT POINT SELECT ============================================================ 1. Regression begins immediately at 12.0.0 Every thread count shows a drop. 2. Regression deepens as versions increase The slope is monotonic: 12.0.x → 12.1.x → 12.2.x → 12.3.x 3. Worst zone is 64–128 threads This is where handler/mutex/allocator changes show up. 4. 12.3.1 is the worst performer at high concurrency -5.38% at 128 threads -2.64% at 256 threads 5. Only anomaly: 12.3.1 at 16 threads is +1.71% — likely noise or a micro‑path improvement that doesn’t scale. Threads | 12.0.0 | 12.0.1 | 12.1.2 | 12.2.1 | 12.3.1 --------+-------------+-------------+-------------+-------------+------------- 16 | -1.98% | -1.13% | -1.77% | -1.73% | +1.71% 32 | -1.40% | -0.37% | -1.19% | -1.84% | -1.08% 64 | -1.74% | -1.64% | -1.84% | -2.54% | -3.19% 128 | -2.14% | -3.23% | -2.86% | -2.89% | -5.38% 256 | -0.99% | -0.81% | -0.50% | -1.45% | -2.64% ============================================================ INTERPRETATION SNAPSHOT — SELECT_SIMPLE_RANGES ============================================================ 1. Regression begins immediately at 12.0.0 Every thread count except one (256 threads, +0.43%) shows a drop. The pattern matches POINT_SELECT: the decline starts right away. 2. Regression deepens as versions increase The slope is monotonic: 12.0.x → 12.1.x → 12.2.x → 12.3.x Same ordering as POINT_SELECT, confirming a shared core-path issue. 3. Worst zone is 32–64 threads This is where the largest drops appear: -3.40% at 32 threads (12.1.2) -2.48% at 64 threads (12.1.2) This is the classic range-select sensitivity zone. 4. 12.3.1 is consistently the worst performer at high concurrency -2.50% at 128 threads -1.64% at 256 threads The degradation grows as concurrency increases. 5. Only anomaly: 12.0.0 at 256 threads is +0.43% — likely noise or a micro-path quirk. Threads | 12.0.0 | 12.0.1 | 12.1.2 | 12.2.1 | 12.3.1 --------+-------------+-------------+-------------+-------------+------------- 16 | -1.34% | -1.69% | -2.79% | -1.20% | -0.06% 32 | -0.80% | -1.45% | -3.40% | -1.25% | -2.20% 64 | -0.88% | -1.59% | -2.48% | -1.15% | -2.04% 128 | -0.14% | -1.75% | -1.57% | -1.94% | -2.50% 256 | +0.43% | -0.96% | -0.88% | -1.15% | -1.64% ============================================================ POINT_SELECT — Worst‑Case TPS Loss ============================================================ Threads | 11.8.6 TPS | Worst TPS (12.x) | TPS Lost --------+------------+------------------+----------- 128 | 370188.986 | 350259.078 | 19929.908 ============================================================ SIMPLE_RANGES — Worst‑Case TPS Loss ============================================================ Threads | 11.8.6 TPS | Worst TPS (12.x) | TPS Lost --------+------------+------------------+----------- 32 | 144376.544 | 139469.276 | 4907.268 --------------------------------------------------------------- Sum and Distinct also showeds possible issues. Here are results of detailed reruns. ============================================================ INTERPRETATION SNAPSHOT — SELECT_SUM_RANGES ============================================================ 1. Regression begins immediately at 12.0.0 Every thread count shows a drop except 16 threads in 12.3.1 (+1.02%). The decline starts right away, consistent with PS and SIMPLE. 2. Regression deepens as versions increase The slope is monotonic: 12.0.x → 12.1.x → 12.2.x → 12.3.x The pattern matches the earlier workloads, confirming a shared core-path issue. 3. Worst zone is 32 threads This is where SUM_RANGES is most sensitive: -4.51% at 32 threads (12.1.2) -5.03% at 32 threads (12.3.1) This mirrors SIMPLE’s behavior, where 32 threads was also the largest drop. 4. 12.3.1 is the worst performer at high concurrency -2.72% at 128 threads -2.52% at 256 threads The degradation grows as concurrency increases, same as PS and SIMPLE. 5. Only anomaly: 12.3.1 at 16 threads is +1.02% — likely noise or a micro-path quirk. It does not alter the overall regression pattern. Threads | 12.0.0 | 12.0.1 | 12.1.2 | 12.2.1 | 12.3.1 --------+-------------+-------------+-------------+-------------+------------- 16 | -1.11% | -1.88% | -2.38% | -2.09% | +1.02% 32 | -2.15% | -2.77% | -4.51% | -1.81% | -5.03% 64 | -0.40% | -0.86% | -0.83% | -0.40% | -0.68% 128 | -0.74% | -1.09% | -1.75% | -1.18% | -2.72% 256 | -0.96% | -1.70% | -2.37% | -1.21% | -2.52% ============================================================ INTERPRETATION SNAPSHOT — SELECT_DISTINCT_RANGES ============================================================ 1. Regression begins at 12.0.1 12.0.0 shows slight improvement at low threads (+1.09%), but the decline starts immediately after. Distinct is more volatile than prior workloads. 2. Regression deepens through 12.1.x and 12.3.x The slope is monotonic overall: 12.0.x → 12.1.x → 12.2.x → 12.3.x The pattern aligns with PS, SIMPLE, and SUM, confirming systemic degradation. 3. Worst zone is 16–64 threads Distinct shows its largest drops early: -4.33% at 16 threads (12.3.1) -3.13% at 32 threads (12.1.2) -3.06% at 64 threads (12.1.2) Distinct is more sensitive to mid‑range concurrency. 4. 12.3.1 is again the worst performer -4.33% at 16 threads -1.89% at 128 threads -1.29% at 256 threads The degradation persists across all concurrency levels. 5. Only anomaly: 12.2.1 shows mild positive deltas at several thread counts (+0.45%, +0.16%, +0.07%, +0.38%). Likely optimizer noise or micro‑path variance, not a structural improvement. Threads | 12.0.0 | 12.0.1 | 12.1.2 | 12.2.1 | 12.3.1 --------+-------------+-------------+-------------+-------------+------------- 16 | +1.09% | -0.70% | -2.21% | -0.56% | -4.33% 32 | +0.93% | -0.82% | -3.13% | +0.45% | -1.54% 64 | +0.99% | -0.96% | -3.06% | +0.16% | -2.17% 128 | +0.17% | -1.22% | -2.73% | +0.07% | -1.89% 256 | +0.50% | -0.87% | -2.20% | +0.38% | -1.29%