Virtualization Technology News and Information
JuiceFS: A New Approach to Data-Intensive File Storage Showcased at the IT Press Tour

As part of the 56th edition of the IT Press Tour event held in California this week, VMblog had the opportunity to attend a briefing by Rui Su, co-founder of Juicedata, the company behind the innovative JuiceFS distributed file system. Su provided an in-depth look at how JuiceFS addresses the limitations of traditional storage solutions when dealing with demanding data-intensive workloads.

The company was founded in 2017 by Davies Liu and Rui Su. Liu is a seasoned developer with contributions to projects like MooseFS, BeansDB, and DPark, and experience at companies like Meta and Databricks. Su brings a diverse background, having been a startup founder, tech lead, and engineer at companies like Douban Movie and Maxthon Browser.

The Need for an Elastic, High-Performance File System

Su began by highlighting the persistent challenges organizations face with existing file systems and object stores like S3. While convenient for web services, these solutions often lack the requisite POSIX compatibility, elasticity, and high throughput to efficiently handle AI/ML, big data, and other data-intensive use cases.

Key pain points include the inability to scale performance independently from capacity, inefficient management of small files so prevalent in the AI domain, and restricted data access across multi-cloud and hybrid environments. These bottlenecks, according to Su, have held back enterprises from fully realizing the potential of their data-driven initiatives.

The JuiceFS Architecture: Designed for Massive Data and High Performance

To tackle these challenges head-on, Juicedata has developed JuiceFS, a POSIX-compliant distributed file system with strong consistency guarantees and support for thousands of concurrent clients and mixed workloads. At its core is a novel architecture that disaggregates storage performance from capacity while enabling independent scaling of throughput.

"With JuiceFS, you can leverage a multi-layer caching strategy, utilizing local storage on compute nodes, as well as dedicated cache pools, to accelerate performance," Su explained. "Meanwhile, an object store backend ensures data durability, allowing you to decouple these critical aspects cost-effectively."

Another standout feature is JuiceFS' native support for multi-cloud and hybrid cloud environments through transparent data replication across regions and on-premises deployments. This flexibility, Su emphasized, is crucial in an era where data and compute resources are increasingly distributed.

Community and Enterprise Flavors for Diverse Needs

During the briefing, Su discussed JuiceFS' availability in both Community (open source) and Enterprise editions, tailored to different user requirements.

The Community Edition, with over 10,000 GitHub stars, targets general-purpose distributed file system needs, emphasizing ease of use, maintenance, and customization. In contrast, the Enterprise Edition is optimized for the most demanding data-intensive workloads, offering advanced capabilities like intelligent caching and granular performance tuning.

Proven Success Across Data-Intensive Industries

Su shared several compelling case studies that demonstrate JuiceFS' real-world impact across sectors like generative AI, autonomous driving, quantitative trading, and biotechnology.

One LLM (Large Language Model) startup achieved over 300 GBps throughput on a hybrid cloud setup for model pre-training using JuiceFS. An autonomous driving company relies on JuiceFS to handle 20 billion files, sustaining 450,000 metadata operations per second and 70 GiBps peak throughput with sub-millisecond latencies.

A biotech leader leveraging JuiceFS for genomic pipelines reported 75% improved performance and 60% cost reduction compared to their previous solution.

Performance and Cost Advantages

Benchmarks and customer deployments position JuiceFS as a high-performance, cost-effective alternative to traditional offerings like S3FS/GoofyS, AWS EFS, and AWS FSx for Lustre. According to Su, while AWS' managed Lustre service can cost upwards of $0.60/GB-month, excluding metadata and data transfer fees, JuiceFS' pricing starts at just $0.04/GB-month through a combination of the Juicedata service fee and underlying cloud storage costs.

Perhaps more importantly, JuiceFS offers unlimited throughput scalability, simply by adding more caching resources, whereas Lustre's aggregate throughput is capped at 1000 MB/s per tebibyte of provisioned storage.

The Road Ahead

As the briefing concluded, Su outlined Juicedata's roadmap, including continuous investment in R&D, growing the JuiceFS community presence in North America, and building out dedicated marketing and sales capabilities to better serve enterprises globally.

With its unique architecture, proven performance benefits, and a flourishing open-source community, JuiceFS from Juicedata is emerging as a compelling solution for organizations seeking to unlock the full potential of their data-intensive workloads across AI, big data, and beyond.


Published Thursday, June 13, 2024 7:25 PM by David Marshall
Filed under: ,
There are no comments for this post.
To post a comment, you must be a registered user. Registration is free and easy! Sign up now!
<June 2024>