PostgreSQL's parallel query planner charges two fees for going parallel: a setup cost per Gather node and a per-row cost for moving data from workers to… (I consult on PostgreSQL through PGX Inc..)

Parallel scans need a minimum table or index size to be considered—but that minimum also sets the bottom rung of the ladder the planner climbs to decide how…

Parallel apply workers speed up logical replication of large transactions—but only if you have enough of them.

Neither of these parameters means what its name says. max_parallel_workers_per_gather is the most workers the planner is allowed to request for one plan node, and the planner usually requests...

max_parallel_maintenance_workers is a ceiling, and most of the time something lower gets there first. It caps the number of parallel workers a single utility command may start. The default is 2,...

Parallel hash joins pool worker memory to build one shared table instead of having each worker build its own copy—a distinction that matters enormously on…

Parallel Append spreads workers across partitions or UNION branches simultaneously, not sequentially.

pgvector 0.8.2 fixes CVE-2026-3172, a heap buffer overflow in parallel HNSW index builds that can leak data or crash the server.

使用GNU parallel加速rsync迁移大量小文件，通过并行处理顶层子目录，减少文件系统元数据操作时间，提高磁盘利用率。注意调整并行进程数以避免机械硬盘过载，迁移后可用rsync校验差异。

PostgreSQL 19 ships with parallel autovacuum. The new GUC autovacuum_max_parallel_workers caps the cluster-wide pool, and the per-table storage parameter autovacuum_parallel_workers lets you tune...

GNU Parallel 是一款强大的命令行工具，能够在 Linux 系统中并行处理任务，提升效率，适用于文件转换和日志分析等场景。它支持多核 CPU 和远程主机，安装简单，语法灵活，适合批量处理和自动化脚本加速。

Logical replication in PostgreSQL has steadily evolved since its introduction in version 10. In a previous blog post, I discussed how PostgreSQL 14 introduced streaming of large transactions....

vLLM 是一款加速大语言模型推理的框架，解决了内存管理瓶颈，实现了 KV 缓存内存零浪费。它支持单节点和多节点的数据并行推理，用户可通过命令行参数配置模型和节点信息，并提供示例代码以便快速上手。

本研究引入TED2025多向平行语料库，解决了多语言低资源环境中未对齐数据的局限性，提升了大型语言模型的多语言表现。实验结果表明，基于多向平行数据训练的模型优于未对齐数据模型。

本研究提出了一种投机性MoE方法，旨在提高大规模混合专家推理的通信效率。通过投机性标记洗牌和专家预调度，显著降低了通信开销，提升了推理效率。实验结果表明，该方法有效改善了DeepSpeed-MoE框架的性能。