Many applications have strict requirements for storage bandwidth and latency. Some applications can’t migrate to Generation 1 clouds because those clouds don’t provide the throughput and price performance of an on-premises file server, but Oracle Cloud Infrastructure can.
Oracle collaborated with IBM preferred partner Re-Store to provide high-performance file servers at scale on Oracle Cloud Infrastructure. This solution meets the needs of HPC workloads like ML/AI, Electronic Design Automation (EDA), and Oil and Gas applications, all of which see throughput as a common bottleneck. This solution architecture deploys in minutes, scales easily, costs a few cents per gigabyte per month, and provides performance that’s consistent with on-premises HPC file servers.
Working with IBM Spectrum Scale, our solution achieves performance that hasn’t previously been seen in the cloud. The following graph shows the throughput achieved by different-size file servers. For example, on a configuration as small as two nodes, a file server on Oracle Cloud Infrastructure provides over 5 GB/s throughput. The graph below shows the result of running industry recognized IOR performance benchmark.
Figure 1: Aggregated Storage Throughput Versus Number of File Servers
We can provide this level of performance because our block storage is backed by NVMe SSD media and our data centers have a fast, flat network architecture. We believe so strongly in the capabilities of our storage offering that our block storage performance is backed by a unique performance SLA. Throughput varies by the size of the volume, and all volumes 1 TB and above provide the maximum 480 MB/s on the balanced performance tier, by default and at no additional charge.
The impact of Oracle Cloud Infrastructure’s performance and scalability directly impacts the bottom line for file servers. Oracle Cloud’s block storage design allow us to attach the same blocks to multiple nodes simultaneously which provides redundancy with no additional effort, decreasing the solution cost of HPC File Servers on Oracle Cloud. As you increase storage capacity and throughput, cost/GB/month decreases.
Figure 2: Cost per GB Goes Down as You Scale Up the Size of the Cluster
“The price performance of IBM Spectrum Scale makes it possible for automotive and financial industry workloads to get the same data performance in the cloud that they have on-premises. We can address workloads that were previously un-addressable in the cloud because of performance and cost barriers.” —Tejas Karmarkar, Senior Manager of Product Management, Oracle
Oracle Cloud Infrastructure is the only public cloud that supports multiple instances attached to the same storage volume for both read and write operations. This is a powerful feature that allows customers to increase their performance and reliability through parallelizing storage operations. Contact us for more information about how to use “multi-attach” blocks in your cloud solutions.
Figure 3: IBM Spectrum Scale NSD Server Model Architecture
Working closely with IBM’s preferred partner, Re-Store, we’ve created a Terraform deployment that lets you automatically deploy in minutes. Customers are already using this architecture with solid benefits, including a leading Oil and Gas company running GPU’s for seismic processing at petabyte scale, and multiple Media and Entertainment customers for ML/AI applications.
“Oracle Cloud is the only major cloud provider delivering Spectrum Scale in a shared-storage architecture on bare-metal hardware. The Oracle offering has been tuned for the demands of HPC and big-data analytics and offers 10X performance advantage over virtualized cloud offerings, improved flexibility, scalability, and manageability.” —Michael Sedlmayer, President, Re-Store
Every use case is different. The only way to know if Oracle Cloud Infrastructure is right for you is to try it. To try, you can select either Oracle Cloud Always Free services or 30-day free trial that includes US$300 in credit to get you started with a range of services, including compute, storage, and network. Also, visit us at booth 609 at SuperComputing 2019 conference, November 17–22, in Denver, Colorado.