If you follow me, you know I’m all about driving improvement in data center environments. So I was happy to see Allison Goodman’s presentation on Driving Innovation in Cloud Storage with Intel® Optane™ Technology on Cloud Field Day because using big data effectively to capture insights is ultimately about high-performance data storage and retrieval.
Data scientists want to capture and analyze big data quickly to glean insights for their business. Time is money. The cloud architect must weigh their requirements against cost, redundancy, and the effectiveness of remote storage. Furthermore, cloud architects need to design the infrastructure so that remote storage appears fast and local while problem-solving and planning for peaks in demand. Typically, local compute storage is sized based on the highest potential need of a single server or application. This results in only 15-25% utilization on a deployment and wasted capacity during normal average operations.
How can you improve resource utilization? Disaggregation
To scale efficiently and improve costs, cloud architects are turning to disaggregated storage. Solid State Drives (SSDs) are hosted on a separate server and assigned to a compute server on demand. The process involves pooling resources, separating storage from compute services, aggregating storage together, and then scaling compute nodes and storage nodes independently. The following image demonstrates this process.
Increased demand for network speed introduces a new bottleneck
Network speeds in data centers are accelerating and expected to reach 100+ Gigabit Ethernet (100GbE) by 2023. While disaggregating storage from compute resources can increase utilization and lower infrastructure costs, increased demand for network speed introduces a new issue—network latency.
Cloud Service Providers (CSPs) plan for up to 90% utilization of storage node network interface card (NIC) bandwidth to deal with random writes. The transition to 100+GbE networks and peripheral component interconnect express (PCIe) Gen 4 SSDs creates a new write buffer bottleneck that NAND flash memory and Non-Volatile Dual In-Line Memory Modules (NVDIMMs) cannot handle efficiently.
You don’t want to move the bottleneck to storage devices. That’s where Intel Optane solutions come in. Optane technology can resolve issues with network latency that other solutions can’t handle.
Intel Optane solutions – bridging critical gaps in the storage and memory hierarchy
Intel Optane products deliver persistent memory, large memory pools, fast caching and fast storage to optimize, store, and move larger, more complicated data sets. They provide high speed and capacity to eliminate processing bottlenecks and improve performance for big data, cloud, storage, high performance computing (HPC), and other demanding applications.
Intel Optane Persistent Memory (PMem) is in a dual in-line memory module (DIMM) package, operates on the dynamic random-access memory (DRAM) bus and can be used as volatile or persistent memory. It comes in two modes – memory mode which delivers affordable memory capacity expansion, and app-direct mode for performance optimization and support for data persistence.
Intel Optane SSDs deliver fast, consistent storage with high endurance. Optane SSDs operate like other storage on the PCIe bus using the nonvolatile memory express (NVMe) protocol. Use this product to help eliminate data center storage bottlenecks and allow for bigger, more affordable data sets.
What’s the best way to use this new technology? It depends
Using Intel Optane technology to implement disaggregated storage depends on your architectural requirements, resources, and capabilities. Tencent Cloud, a major CSP with in-depth expertise and experience, developed their Cloud Block Storage (CBS) service using Intel Optane PMem, customized 2nd Gen Intel® Xeon® Scalable processors, and Intel® Deep Learning Boost (Intel® DL Boost).
The CBS solution provides highly efficient and reliable persistent block storage to support cloud workloads for Tencent Cloud’s customers, including high-load OLTP financial transaction systems, high-throughput e-commerce systems, data analytics systems for artificial intelligence (AI), and high-concurrency content delivery networks. It provides bandwidth capability up to 4Gbps and 1.1 million IOPs on a single disk of up to 32 TB, with support for a single cloud host to scale up to 640 TB.
If you don’t have in-house developer resources, you can partner with an Independent Software Vendor (ISV) such as Lightbits Labs. Lightbits’ LightOS service is a composable disaggregated software-defined storage solution that pools NVMe storage across multiple storage devices into a high-performance virtual all-flash storage array. LightOS uses Intel Optane PMem to drive large volumes of memory at a lower cost than DRAM, which is important for metadata. Optane DIMMs make the array operations faster than DRAM, while Intel® 3D NAND SSDs based on Intel® QLC Technology increase capacity and decrease cost/GB compared to DRAM.
Hardware choices form the foundation for innovation
Moving to disaggregated storage delivers multiple benefits and to do it well, you need the speed to move data across the network as quickly as possible. Cloud architects can work with Intel Optane PMem and Intel Optane SSDs to resolve issues with network bottlenecks.
When fast 100+GbE networks are coupled with fast storage node compute and Intel Optane technology, remote storage can appear as local storage to applications, which makes data scientists and other cloud users very happy. Solutions can still scale out capacity independently from compute, allowing for much better resource utilization. It’s a win-win for everyone.
Innovative technology, adaptable to your needs across workloads
Intel Optane PMem and Optane SSDs address the performance and capacity gap between DRAM and NAND storage. Intel Optane PMem provides a unique combination of larger capacity and support for data persistence. You can affordably increase your memory capacity per socket, extract value from larger data sets and increase the utility of each server. Intel Optane SSDs provide high throughput for breakthrough performance. You can accelerate applications, reduce transaction costs for latency-sensitive workloads, and improve overall data center TCO.
To learn how Intel Optane technology delivers high speed and density, eliminates processing bottlenecks, and improves performance, check out the following resources:
- Distributed Storage Trends and Implications for Cloud Storage Planners
- Intel Optane Persistent Memory Accelerates Scientific Computing
Notices & Disclaimers
Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex. Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available updates. No product or component can be absolutely secure. Your costs and results may vary. Intel technologies may require enabled hardware, software or service activation.