blog

Optimising OCI with RDMA

Written by Ian Rogers | Apr 26, 2024 1:08:41 AM

2 Min Read

Optimising Oracle Cloud Infrastructure (OCI) with Remote Direct Memory Access (RDMA)

In sectors like banking, finance, and global investments, where microseconds matter, the demand for swift transactions is insatiable. The surge in Machine Learning models and AI has only heightened this need for speed.

Speed, in this context, is often quantified by latency, where lower latency signifies superior performance, ideally aiming for zero latency.

RDMA empowers networked computers to exchange data in main memory without involving their CPUs, caches, or operating systems. This results in high-throughput, low-latency networking, especially beneficial for massive parallel computer clusters.

Once confined to supercomputers and large data centres, RDMA technology has evolved significantly since its inception. Today, OCI harnesses RDMA to optimise enterprise computing performance and spearhead advancements in artificial intelligence (AI).

Oracle's acquisition of Sun Microsystems provided access to hardware engineering for high-performance computing, including Mellanox cards and InfiniBand technology. Collaborating with Mellanox (now under Nvidia's umbrella), Oracle developed RDMA over RoCE cluster networking, a distinguishing factor for Exadata and HPC on OCI. Larry Ellison, founder and CTO, has described RDMA as part of OCI's secret sauce, enabling superior performance at a reduced cost.

RDMA facilitates direct data exchange in main memory between networked computers, bypassing the processor, cache, or operating system of either machine. This optimisation enhances throughput and performance, resulting in faster data transfer rates and reduced latency between RDMA-enabled systems. OCI's cluster networking now supports RDMA-connected clusters of up to 20,000 cores, delivering low-latency networking and application scalability similar to on-premises clusters.

On an Oracle Q2 2024 earnings call in December 2023, founder and CTO Larry Ellison said the company is in the process of expanding 67 of its existing cloud regions and building 100 new cloud data centres. Larry Ellison added “We're able to build new data centres rapidly and operate them inexpensively because all of our data centres are architecturally identical, highly automated, with an identical high-performance RDMA network, autonomous services, and applications.”

TEAM Cloud, New Zealand's forthcoming hyperscale sovereign cloud service operated by TEAM IM and powered by OCI with RDMA-enabled cluster networking, epitomises this technological synergy. By leveraging OCI's capabilities, TEAM Cloud will deliver high performance and low latency for enterprise applications from its cloud regions.

 

Author

Ian Rogers - TEAM Cloud Founder & CEO at TEAM IM