This blog is written by Tony Ansley & Larry Hart.
At Micron, our focus is to deliver high performance solutions to our customers for their business needs while maximizing their budget. To achieve this, we are always looking for ways to create new and innovative solutions using the latest technology trends to help our customers create high-performance solutions for their business needs allowing you to maximize your budget while using Micron SSDs. Server SoCs based on Arm is one technology trend that shows promise and that uses advanced storage technologies right out of the gate. When paired with advanced DRAM and SSDs from Micron, you can realize impressive performance results. These CPUs may also very well-suited for use in advanced scale-out, software-defined storage (SDS) solutions such as Excelero NVMesh® or E8 Storage®.
How do we know? Because we have seen the results first hand. We teamed up with our friends at Marvell® (Marvell acquired Cavium®™ in July of 2018) to perform some typical cloud-centric application benchmarks using their Armv8-based ThunderX2® reference server platform with Micron’s latest generation NVMe-based 9200 SSDs. Micron’s 9200 series of SSDs provides high-performance PCIe® interfaces that ensure high throughput and low latency along with advanced enterprise security.
So what solutions did we investigate and what were the results? We evaluated two different workloads, first a cloud-ready NoSQL database solution represented by Apache Cassandra® and then an open-source scale-out SDS solution based on the community edition of Red Hat® Ceph™ Luminous 12.2.41 with BlueStore. Both tests prove that Marvell and Micron can provide a flexible, high-performance, all-flash server using open-source, cloud-ready storage solutions on an advanced ARM-based server platform.
Both benchmarks used a basic 4-node cluster for the data engine components connected to a 100Gb Ethernet network.
Apache Cassandra® for Faster Insights
Using the popular Yahoo! Cloud Service Benchmark (YCSB) suite, the Cassandra tests focused on read-centric workloads that represent typical data analytic solutions. Workloads B, C and D were tested2. The Arm-based servers generated over 400,000 operations per second for workloads B and C and over 100,000 operations per second for workload D. This performance is better than our previous test results with x86 CPU-based server solutions3. More details about this Marvell and Micron solution are available in the Cassandra solution brief.
Ceph Object Stores Get a Boost
Our second benchmark looked at the ground-breaking open-source SDS solution Ceph. Used in many complex, scalable, private cloud and big-data solutions, Ceph is a high-performance storage solution that supports block and object content types. With its newest feature addition called Bluestore, Ceph can now be optimized for flash storage solutions, such as Micron’s 9200 NVMe SSDs. To perform this test, we used the Ceph RADOS Bench tool to simulate 100% read and 100% write traffic using a 4MB object-size. The results were excellent, with 45GB/s read throughput and 10GB/s of write throughput. This is similar to earlier Ceph results using x86 CPU-based server solutions4. For more details about this Marvell and Micron solution, review the Ceph solution brief.
Clearly, Arm-based SoC platforms are a viable option for scalable, cloud-centric workloads. Whether you are considering NoSQL solutions for your data analytics, high-performance cloud data repositories, or other workloads, Micron’s 9200 NVMe SSDs hosted by Marvell's Arm-based ThunderX2 server architecture are up to the challenge. If you are considering Arm®-based technology as an alternative for your cloud-centric solution, be sure to contact your Micron channel representative for all the details.
1. Red Hat does not endorse any of the stated tests or results discussed in this blog.
2. YCSB Workloads B,C, and D represent typical read-centric workloads that are typical of data analytics solutions. Workloads A, E, and F are write-centric workloads.