All-flash storage: Faster, more stable, more cost-effective

Isn't it time you finally bid farewell to the era of mechanical data vaults?

Huawei in September 2019 launched its latest all-flash storage array, the OceanStor Dorado, a system that delivers the next-generation of intelligent storage.

The Oceanstor Dorado performs up to 20 million I/O operations per second (IOPS) – twice as much as the next-best player. Further, it features an AI processor, a first for a storage array and something that enables performance tuning and management of the system. The OceanStor Dorado also provides fault-tolerance features and a one-second switchover with uninterrupted links in the event of controller failure.

Huawei has invested in the development of flash storage for 17 years with the result that more than 10,000 customers worldwide have chosen Huawei's storage products. Huawei OceanStor Dorado has been deployed in the core systems of organisations in different industrial sectors including finance, telecommunication and energy.

This success in the storage and all-flash markets has allowed Huawei to not just maintain a lead in market share domestically but to also hit a high level of growth globally.

Huawei plans to support the continued and widespread adoption of all-flash storage with advances in the OceanStor Dorado that make it an even faster, more stable, and more cost-effective all-flash storage, and that allow more enterprises to finally bid farewell to the mechanical era of storage.

Welcome to the family

There are five OceanStor Dorado models, which all comprise a multi-controller, fully-interconnected and shared architecture design on the front and back end. The front-end is connected through Fibre Channel (FC), and the back-end through 100 Gigabit Ethernet (100GE) RoCE. Furthermore, the storage system is divided into different product types according to the number of controllers, controller cache, number of front-end interfaces, as well as the maximum number of hot-swappable I/O modules.

The latest OceanStor Dorado is designed so that the front-end interconnect I/O module, controller, and back-end interconnect I/O module are fully connected at all three layers to ensure high reliability and performance. In the event of a failure, a controller switchover can be performed within one second, while host links remain uninterrupted and upper-layer services unaffected. The central processing units (CPUs) of the controllers and the solid-state drive (SSD) disk controllers incorporate Huawei’s Kunpeng 920 microprocessor chip in addition to Huawei Ascend chips for advanced AI.

Enterprises increasingly demand improvements in storage input/output operations per second (IOPS) and latency to ensure highly reliable services for key business applications and systems, such as core banking transaction systems or telecom charging systems. In terms of evaluating storage products, most enterprises focus on the IOPS as a key indicator of performance, much like the Linkpack test in high-performance computing (HPC). For online transaction applications such as databases and online transaction processing (OLTP), however, a high IOPS does not necessarily ensure a strong transaction processing system (TPS), rather it is latency that makes the difference. Hence, the saying: "IOPS is nothing without latency".

To meet these requirements, the performance of the latest OceanStor Dorado has been turned to exceed the previous-generation product, reaching 20 million IOPS and 0.1 ms latency. This improved performance is attributable to the Kunpeng 920 chip's multi-core processing capability, the industry-leading CoreFarm intelligent scheduling algorithm, as well as the self-acceleration of the data storage driven by the Kunpeng A310 smart chip, particularly for latency control.

In terms of reliability, the most frequently occurring error in the storage system is a failure in the storage controller. Generally, storage systems can manage two simultaneous failure in controllers, but is this sufficient? It may not be adequate in a scenario where an entire controller enclosure experiences failure or shuts down suddenly which often involves the simultaneous failure of multiple controllers. When this happens, it’s crucial that the customer's core service system is not interrupted. With an architecture designed for reliability, Huawei's storage systems can manage up to seven out of eight controller failures simultaneously, allowing service systems to operate uninterrupted, effectively eradicating the occurrence of full controller enclosure failures.

Upgrades no problem

Another requirement in the enterprises is to minimize the business impact of system upgrades during the storage lifecycle. The industry’s fastest system upgrade can be completed within two hours, from pre-upgrade inspections, execution of the upgrade to post-update inspection, and any interruption to services can be resolved within 10 seconds. Huawei, however, has taken this to a new level, completing the upgrade within 30 minutes and resolving service interruptions within one second. This exceptional reliability is made possible through Huawei’s SmartMatrix full-mesh architecture that connects the I/O interface module on the front and back end, controllers, and the back-end disk enclosures through the utilization of the remote direct memory access (RDMA) high-speed network.

In addition to flash storage systems, Huawei also develops its own SSDs. To accelerate the popularization of flash storage, Huawei has initiated the 1:1 replacement of 10K SAS hard drives with SSDs in China.

Seventeen years on, Huawei is just getting started with the OceanStor Dorado. Huawei will continue to utilize its unique deep-disk controllers to reduce costs and improve the service life and performance of SSDs in the enterprise, while also improving data reduction and further promoting flash storage with a combination of scenario-specific features and high-performance computing power.