Blog
Zilliz Cloud: Igniting Vector Searching with Rocket-Like Speed

Zilliz Cloud: Igniting Vector Searching with Rocket-Like Speed

Jul 19, 20235 min read

We recently launched the newest version of Zilliz Cloud, introducing an array of exciting features, including a new free tier, dynamic schema and partition keys, and more affordable pricing plans to cater to diverse business needs. We've also significantly improved performance, making Zilliz Cloud twice as fast as the previous version and three to ten times faster than Milvus and other vector databases.

As a software engineer, I am particularly excited about Zilliz Cloud's rocket-like speed of handling queries. In this post, I will showcase how fast Zilliz Cloud is and what makes Zilliz lightning fast.

How fast is Zilliz Cloud?

Before demonstrating the speed of Zilliz Cloud, we need to establish the context of our comparison and the specific conditions we're evaluating.

VectorDBBench is an open-source benchmarking tool that evaluates different vector databases' performance across multiple metrics, such as QPS, capacity, throughput, and latency, through an intuitive and user-friendly interface.

I'll use VectorDBBench to compare Zilliz Cloud with Milvus, an open-source database created by Zilliz, and other vector databases in the market, including Quadrant Cloud, Pinecone, WeaviateCloud, and ElasticCloud.

Zilliz Cloud easily outperforms Milvus by 6x

The figure below compares the performance of Zilliz Cloud and Milvus in terms of QPS when handling one million 768-dimensional data with different compute and storage resource configurations. Based on the results, Zilliz Cloud is at least three times faster than Milvus.

Zilliz Cloud vs. Milvus (1M dataset, 768 Dim)

When retrieving 10 million 768-dimensional data, Zilliz Cloud outperforms Milvus by six times.

Zilliz Cloud vs. Milvus (10M dataset, 768 Dim)

Zilliz Cloud is at least 4x faster than other vector databases

Zilliz Cloud is also faster than other vector databases. Based on the benchmarking results, Zilliz Cloud achieves 4 to 191 times higher QPS than competitors when retrieving one million 768-dimensional data and offers much more affordable pricing plans.

Zilliz Cloud vs. other vector databases (1M dataset, 768 Dim)

When performing similarity searches on larger datasets, Zilliz Cloud surpasses vector database services such as Pinecone and Quadrant Cloud by 3x regarding QPS.

Zilliz Cloud vs. other vector databases (10M dataset, 768 Dim)

For more results, check our benchmarking page.

Why Zilliz Cloud is so fast

Now you’ve read about the speed of Zilliz Cloud; it’s time to uncover what makes it so fast.

Zilliz Cloud has a robust vector indexing engine

Vector databases rely heavily on computing power, with their vector indexing algorithms consuming most resources and mainly determining the performance.

Zilliz Cloud has a robust indexing engine with powerful vector computing capabilities. Glass is the open-source version and topped the rankings in recent test results released by the ANN-Benchmark. This well-known benchmarking tool compares the performance of different algorithms on various real datasets.

Let’s take a look at two examples. The following diagrams demonstrate ANN-Benchmark’s most recent benchmarking results based on two different datasets:

gist-960-euclidean: one million 960-dimensional vectors using the euclidean distance function.
fashion-mnist-784-euclidean: 60,000 784-dimensional vectors using the euclidean distance function.

The higher the curve reaches, the more queries the algorithms can process per second. Similarly, the more the curve stretches to the right, the higher recall the algorithms can get.

ANN Benchmark results (dataset: gist-960-euclidean, k=10)

Based on the results shown in the diagrams, Glass (the pink curve) ranked as the top performer in terms of both queries per second and recall. Glass's exceptional performance has contributed significantly to Zilliz Cloud's ability to achieve lightning-fast speeds.

Optimized code structure

In addition to Glass's robust indexing capability, we have optimized Zilliz Cloud's code structure. This optimization includes cleaning up redundant query paths, improving the scheduling of query merges and concurrent query rules, optimizing the load balancing structure, and replacing/upgrading many inefficient and outdated third-party libraries. All these efforts help Zilliz Cloud maximize its index algorithm capabilities and reduce overhead.

We have also addressed several issues that previously hindered Zilliz Cloud's query performance. As a result, scalar-filtered and vector-filtered searches with large-scale datasets are now significantly faster.

Zilliz Cloud’s AutoIndex for stable recall rates

The recall rate is a critical factor for system performance. An inadequate recall rate will lead to retrieval results that do not meet our customer’s requirements. However, an excessively high recall rate may significantly compromise performance. Therefore, we need a stable recall rate that balances performance and accuracy, ensuring predictable results and better support for various production scenarios.

Controlling the recall rate is essential but also complicated because the recall rate can be affected by various factors, including the data size, indexing and query parameters, data distribution, and even the size of topk results. Most database systems use one of the following two solutions to address this problem:

Allow users to provide parameters: flexible but challenging to adopt.
Provide empirical parameters: easy to use but less flexible and cannot precisely control situations with different datasets and topk requirements.

Zilliz Cloud offers a more straightforward and flexible solution called AutoIndex. It analyzes practical application cases based on models and customizes parameters for each indexing request. This approach ensures a stable recall rate while delivering excellent performance to meet users' requirements in practical applications.

Summary

The latest update of Zilliz Cloud has made milestone improvements in functionality and performance. It now boasts rocket-like speed in handling massive vector searches, significantly outperforming other vector databases.

Getting started with Zilliz Cloud

Start for free with the new Starter Plan! A great plan to get started with no installation hassles and does not require a credit card!
Or start your 30-day free trial of the Standard plan with $100 worth of credits upon registration and the opportunity to earn up to $200 worth of credits in total.
Dive deeper into the Zilliz Cloud documentation.
Check out the guide on migrating from Milvus to Zilliz Cloud.

Li Liu
Li Liu is the Principal Engineer at Zilliz, leading the vector searching research and development. Before joining Zilliz, he was a Senior Engineer at Meta, designing and shaping numerous advertising stream data frameworks. With a Master's degree from Carnegie Mellon University, he boasts extensive experience in databases and big data. Li Liu's expertise in technology and innovation continues to drive advancements in vector searching, leaving a lasting impact on the field.