At Google Cloud Next ’26, Google introduced Cloud Storage Rapid, a new suite of object storage capabilities designed specifically for data-intensive tasks such as artificial intelligence (AI) and analytics. This innovative offering includes Rapid Bucket, a high-performance zonal object storage solution, and Rapid Cache, which enhances read speeds and integrates compute and data for existing workloads.
As organizations increasingly adopt AI technologies, the demand for efficient storage solutions has grown. Storage systems play a crucial role in supporting accelerators like GPUs and TPUs during training and inference processes. However, as workloads scale, traditional storage solutions often become bottlenecks, hindering performance and wasting valuable compute resources.
Rapid Bucket: High-Performance Storage
Rapid Bucket leverages Google’s Colossus distributed storage system to offer exceptional read and write performance with ultra-low latency. Key features include:
- Ultra-low latency: Capable of handling up to 20 million queries per second.
- Massive scalability: Provides over 15 TB/s of read throughput from a single zonal bucket.
- New functionalities: Introduces native appends, unlimited readers during writes, and vectored reads.
Rapid Bucket is optimized for various demanding applications, such as AI/ML data preparation, training, and analytics processing, resulting in:
- Improved accelerator utilization: Reduces blocked GPU time by 50% and accelerates data loading by up to 2.5 times.
- Faster checkpointing: Enables checkpoint restores that are five times quicker and writes that are 3.2 times faster than traditional storage.
Rapid Cache: Accelerating Read Performance
Rapid Cache, which debuted at Cloud Next ‘25, boosts bandwidth for AI/ML workloads without requiring code modifications. It achieves an aggregate read throughput of 2.5 TB/s and offers up to 2.1 times faster model loading, resulting in significant cost savings.
One of the standout features of Rapid Cache is its new ingest-on-write capability, which allows data to be cached immediately as it is written. This reduces initial cache misses and improves checkpoint restore times by up to 2.2 times.
Case Study: Thinking Machines Lab
Thinking Machines Lab, an AI research company, has integrated Rapid Cache into its workflows to address the infrastructure challenges posed by large-scale data processing. By utilizing Rapid Cache, the lab has achieved:
- Consistent high bandwidth: Stable read throughput peaks exceeding 1.8 TB/s.
- Enhanced stability: Significant reduction in latency and errors during high-demand periods.
- Operational efficiency: Improved management of massive Spark workloads through optimized directory handling.
Conclusion: The Future of Cloud Storage
Cloud Storage Rapid represents a significant advancement in storage solutions for AI and analytics. With its high performance and reliability, Rapid Bucket and Rapid Cache are designed to meet the needs of modern data-intensive workloads. Organizations looking to enhance their data processing capabilities can explore these solutions to optimize their operations.
