Disaggregated Systems Driven by Cloud Economics and Network Advances

The Economic Imperative for Disaggregation

Murat Demirbas, a distributed systems expert from MongoDB Research and formerly AWS, explained that cloud economics are the main force behind the adoption of disaggregated systems. The fundamental difference between compute and storage — compute being expensive and fluctuating, while storage is cheap and stable — makes their tight coupling inefficient. When bundled, increasing one often forces an increase in the other, leading to higher costs for customers.

Technical Enablers of Disaggregation

The shift to disaggregated data centers has been facilitated by significant improvements in networking technology. Current networks offer hundreds of gigabytes per second bandwidth, an order of magnitude increase over 10-15 years ago. Advanced networking technologies such as RDMA, SmartNICs, and CXL further support this architectural evolution by providing the necessary speed and efficiency for separated components to communicate effectively.

Benefits for Cloud Providers and Users

Disaggregation allows for independent scaling of compute and storage. Compute can scale up for latency improvements or scale out for increased bandwidth and parallel processing. Critically, it enables scaling compute down to zero, aligning with the pay-per-use model that customers value. Another cost benefit is the ability to implement I/O pooling at the storage tier, optimizing resource utilization.