https://www.professionaldev.dev/tags/jdbc/Recent content in Jdbc on Professional developerHugo -- gohugo.ioenMon, 19 Oct 2026 00:00:00 +0000https://www.professionaldev.dev/post/java/mssqlserver/fast_insert_part8/Mon, 19 Oct 2026 00:00:00 +0000https://www.professionaldev.dev/post/java/mssqlserver/fast_insert_part8/ <h2 id="executive-summary">Executive Summary</h2> <p>This series benchmarks <strong>110+ insert strategies and tuning knobs</strong> on SQL Server 2025 Express (Windows, Java 17, Spring Boot 3.3) over <strong>4,222 individual benchmark runs</strong>. Here's the TL;DR:</p> <h3 id="the-10-highest-impact-decisions-in-order">The 10 highest-impact decisions (in order)</h3> <table> <thead> <tr> <th>#</th> <th>Choice</th> <th>Impact</th> </tr> </thead> <tbody> <tr> <td>1</td> <td><strong>Batch your inserts</strong> (any strategy ≫ auto-commit per row)</td> <td><strong>15x</strong></td> </tr> <tr> <td>2</td> <td><strong>Use the Bulk Copy API</strong> (<code>SQLServerBulkCopy</code> / <code>insertBulk...</code>)</td> <td><strong>5-6x</strong> vs plain JDBC</td> </tr> <tr> <td>3</td> <td><strong>Parallelize across 4 connections</strong></td> <td><strong>2-3x</strong> on top of #2</td> </tr> <tr> <td>4</td> <td><strong>Enable <code>useBulkCopyForBatchInsert=true</code></strong> in the JDBC URL</td> <td><strong>2-2.5x</strong></td> </tr> <tr> <td>5</td> <td><strong>Use multi-value INSERT (200 values/statement)</strong></td> <td><strong>4-5x</strong> for JDBC</td> </tr> <tr> <td>6</td> <td><strong>Drop/disable non-essential indexes</strong> before bulk loads</td> <td>up to <strong>7x</strong></td> </tr> <tr> <td>7</td> <td><strong>Pick the right batch size (500-1000)</strong> — not 10, not 10,000</td> <td><strong>4-5x</strong> vs tiny batches</td> </tr> <tr> <td>8</td> <td><strong>Add <code>WITH (TABLOCK)</code> hint</strong> for dedicated bulk-load windows</td> <td><strong>1.3x</strong></td> </tr> <tr> <td>9</td> <td><strong>Use <code>Persistable</code> + <code>isNew=true</code></strong> if you must use Spring Data <code>saveAll</code></td> <td><strong>8x</strong></td> </tr> <tr> <td>10</td> <td><strong>Keep AUTO_UPDATE_STATISTICS = ON</strong> (don't try to be clever)</td> <td><strong>1.5x</strong></td> </tr> </tbody> </table> <h3 id="top-7-counter-intuitive-findings">Top 7 counter-intuitive findings</h3> <ol> <li><strong><code>OPTION (RECOMPILE)</code> makes batched inserts 52x slower</strong> (2.4 minutes vs 3 seconds).</li> <li><strong>Delayed Durability <code>= FORCED</code> makes bulk inserts 49% SLOWER</strong>, not faster. It helps many-small-commits workloads, not single-large-transaction bulk loads.</li> <li><strong>Natively compiled stored procs are 21x slower</strong> when called from Java per-row via <code>addBatch</code>. Their design assumes the C#/.NET memory-table API.</li> <li><strong>In-Memory OLTP tables don't help batched Java inserts.</strong> They shine for concurrency, not batch throughput.</li> <li><strong>ROW/PAGE compression make inserts slower</strong> (+7%/+33%). Compression is a <em>read</em> optimization.</li> <li><strong>BCP utility is slower than Java Bulk API</strong> when data is in-memory. The file I/O overhead dominates.</li> <li><strong>AUTO_UPDATE_STATISTICS = OFF makes inserts 50% SLOWER</strong>, not faster. The optimizer needs stats even for simple INSERTs.</li> </ol> <h3 id="the-absolute-fastest-paths-if-your-scenario-allows">The absolute fastest paths (if your scenario allows)</h3> <p>If your source data is already in SQL Server (ETL within the same database):</p>