September 2022 Summaries

The latest release of Qdrant 0.10.0 introduces a batch search mechanism that optimizes vector search by allowing multiple search requests to be handled in a single API call, reducing network overhead and improving efficiency. Previously, multiple parallel requests were needed, which could slow down the process, but now Qdrant performs these requests optimally in one call. This feature is particularly beneficial for scenarios requiring multiple non-related tasks to be completed simultaneously. The integration of the official Python client demonstrates the ease of adopting batch search, with step-by-step guidance on creating collections, loading vectors, and executing batch searches. Benchmarks indicate that using batch search can significantly reduce search times, with combinations of parallel processing and batch search showing the greatest time savings. The batch search not only simplifies application code but also enhances performance by potentially reducing query times by over 30%, making it a valuable tool for optimizing vector search tasks.

Qdrant has introduced support for ARM architecture by providing Docker images specifically for ARM users, addressing the growing trend of ARM processors due to their energy efficiency and cost-effectiveness compared to x86 counterparts. This move aligns with Apple's introduction of ARM-based M1 chips and the availability of ARM processors as affordable alternatives in cloud hosting environments. Qdrant conducted performance tests comparing ARM64 and x86 architectures on AWS EC2 instances, revealing that while ARM64 is about 10% slower on average and 20% slower on the median, it offers more consistent performance and is 20% cheaper than x86 options. This makes ARM64 a compelling, cost-effective choice for those looking to implement vector search with Qdrant, despite the slight reduction in speed.

Qdrant 0.10 introduces several enhancements, including the ability to store multiple vectors per object in a single collection, streamlining semantic search and reducing setup complexity. The update also features a new batch vector search capability, allowing multiple search requests to be processed with a single API call, thus minimizing network overhead and improving performance. Additionally, ARM support has been incorporated by providing Docker images specifically compiled for ARM platforms, enhancing accessibility and performance for ARM users. The update also introduces full-text filtering, which complements existing keyword filters, offering users increased flexibility in refining search results on Qdrant's vector database.