Introduction to Cloud ESSD provides cloud server ECS with low-latency, durable, and highly reliable block storage services, becoming the industry benchmark for cloud vendors' all-flash block storage. The storage team launched the new ESSD Auto PL cloud disk specification, which decouples performance and capacity and provides two key features of IO performance on demand. The flexibility and elasticity of AutoPL reduces the difficulty of IT scale planning and the risks caused by improper planning. This article details the features of Auto PL's new products and the technical principles behind the secrets.
Preface
As one of the most important core components of IaaS, Alibaba Cloud ESSD provides cloud server ECS with low-latency, durable, and highly reliable block storage services, becoming the industry benchmark for cloud vendors' all-flash block storage. As more and more enterprises go to the cloud and core applications to the cloud, as well as the vigorous development of container and serverless architectures, new challenges and demands are raised for the elasticity of block storage IO performance. In this context, the Alibaba Cloud storage team launched the new ESSD Auto PL cloud disk specification, which decouples performance and capacity and provides two key features of IO performance on demand. This article combines the typical business scenarios of block storage to introduce the new product features of Auto PL and the technical principles behind the secrets.
IO elastic demand and business pain points of cloud storage
With the development of cloud-native technology, more and more enterprises are building large-scale enterprises based on cloud computing virtualization, elastic expansion, and the booming cloud-native technology distributed framework, container technology, orchestration system, continuous delivery, and rapid iteration. Large-scale, flexible expansion, and rich distributed business scenarios on the cloud; new computing forms are gradually developing in the direction of short cycle, light weight, etc., which puts more demands on the elasticity of block storage IO performance (performance usually uses IOPS: Input/Output Operations per Second and Throughput BPS: Bytes per Second to describe), the following are the more common business pain points:
- VM/container batch startup: When a computing instance is started, the system disk consumes a lot of IOPS and throughput BPS in a short period of time
- Business peak: The customer's business is facing unexpected unexpected scenarios, requiring cloud disks and VMs to have elastic expansion capabilities for short-term sudden performance requirements
- Periodic task processing: OLAP/batch processing periodically submits massive tasks in a foreseeable time, requiring cloud disks with sudden elastic expansion capabilities
Traditional block storage products adopt a performance/capacity coupling product design. Users can obtain the corresponding IOPS/BPS performance upper limit by purchasing cloud disk capacity, and obtain both disk capacity and IO performance through cloud disk expansion. ESSD supports multiple performance levels of PL0/1/2/3 (PL: performance level). Different PL levels have different IO performance upper limits. Customers can increase the PL level through the cloud disk configuration function to obtain higher IOPS/ The upper limit of BPS performance. Cloud-native services make full use of the elasticity of the cloud, and there is a long period of time for business requirements, and some storage performance margins are usually reserved. In addition, a considerable part of the business traffic on the cloud has obvious peaks and troughs, and most of the time is in a period of low business load, and it is difficult to accurately predict the peak period and peak value of the business. A typical IO traffic burst service may have one or more burst IO traffic within a certain period of time. The burst time is short and the burst performance peak is high. It is common in burst service scenarios such as Internet spikes. New performance planning is proposed. Challenge: If the performance configuration reservation is too high, it will cause a lot of idle waste of daily resources; and if the performance reservation is insufficient, the business will be damaged due to sudden floods. All in all, it becomes very difficult to perform more accurate performance planning through cloud disk expansion/configuration changes.
ESSD Auto PL
In response to the above business pain points, Alibaba Cloud has launched the ESSD Auto PL product specification, which supports two modes of performance on-demand configuration and on-demand burst, and supports the ultra-high unit capacity performance upper limit of 1000 IOPS/GB. Performance on-demand configuration is mainly oriented to predictable periodic IO traffic scenarios. In addition to selecting storage capacity when creating ESSD Auto PL, users can also configure additional IO performance upper limits separately to achieve decoupling of IO performance and capacity. In response to the predictable IO peak, users can flexibly adjust IO performance according to business needs and provide predictable response capabilities.
For unexpected business peaks, Auto PL supports performance-on-demand burst mode, providing maximum single disk 100W IOPS, 4GB/s extreme IO performance, cloud disks are automatically adjusted according to actual performance requirements, without IO performance prediction and planning , It makes full use of the elasticity of ESSD distributed storage, and completely solves the problem of performance planning under burst traffic. This function adopts a post-payment model, and users only need to pay for the actual number of reads and writes that exceed the pre-configured performance, which ensures the stable operation of the business and maximizes the saving of the user's resource configuration overhead. Take a burst traffic scenario of a large Internet e-commerce company as an example. The service originally used ESSD PL1, with a performance limit of 50,000 IOPS and 350MB/s. In a business burst traffic scenario, 2.3% of cloud disks have full PL1 performance limit impact Business, and the peak time of the business is relatively short, and the peak traffic cannot be accurately estimated. The traditional need to use ESSD PL2 to meet business burst traffic, use ESSD Auto PL and enable on-demand burst mode, business storage TCO decreased by 49%.
Auto PL is still compatible with the benchmark performance of ESSD PL1. The performance of the standard Auto PL cloud disk is exactly the same as that of ESSD PL1, enabling seamless switching between existing customers and business scenarios. In addition, ESSD Auto PL supports both performance-on-demand configuration and performance-on-demand burst functions for the first time in the industry, and can be superimposed, and users can flexibly configure according to the actual IO traffic model.
Auto PL technology analysis
As the first cloud disk that supports decoupling of performance and capacity and supports elastic scaling of performance according to load, ESSD Auto PL needs to solve many technical challenges: for example, how to quickly sense business load changes, and how to dynamically request and release resources on demand to support performance scaling , How to quickly balance load scheduling and so on. After repeated polishing, ESSD Auto PL cloud disk has designed a fine-grained cloud disk segmentation mechanism, which allows it to use the resources of the entire back-end storage cluster in a balanced manner and quickly and dynamically adjust; through cluster capacity/performance level real-time monitoring and scheduling, multi-level QoS isolation and other issues to guarantee IO performance, such as traffic impact and multi-tenant IO interference caused by sudden bursts of IO performance.
Cloud disk fine-grained segmentation
ESSD Auto PL supports a maximum of 1000 IOPS/GB, far exceeding the IOPS performance per unit capacity of Nand SSD. The LBA address space of each ESSD cloud disk will be divided into multiple stripe groups. The IOs of the stripe groups are broken up by distributed algorithms and processed by different storage nodes to make full use of the RDMA network and high-performance storage capabilities. ESSD Auto PL has designed a fine-grained address space management mechanism, so that small-capacity cloud disks can be fully dispersed to multiple storage nodes, achieving a larger range of IO scheduling capabilities, and a large range of IO scheduling capabilities can also reduce storage clusters Stand-alone hotspots and some IO long-tail delays.
Multi-tenant isolation and IO priority management
EBS is a typical multi-tenant service. The burst of high throughput/high IOPS traffic will potentially affect the IO latency of low-load tenants. The extreme performance of 100W IOPS IO Burst puts forward higher requirements for isolation capabilities. ESSD supports two layers of QoS for instance and cloud disk. Instance QoS provides IO isolation capability between multiple virtual machines. The upper limit is strongly related to the number of vCPU cores of the instance purchased by the user. Some small-sized instances support storage credit burst capability, which can accumulate idle IO quota provides a maximum performance burst capability of 30 minutes; cloud disk QoS provides the upper limit of the performance of each cloud disk in the instance, which is related to the cloud disk specifications. The IO sent from the VM passes through the cloud disk and instance two-level QoS in turn from the link, and performs Burst IO traffic marking to ensure that the entire link can accurately identify Burst traffic in traffic congestion scenarios and ensure that non-Burst traffic is prioritized. Aiming at the system's local hot spots and IO congestion caused by Burst IO traffic, it realizes 10 millisecond level of business load perception and prediction of IO traffic, completes dynamic queue scheduling and concurrency adjustment in seconds, and combines the dynamic queue distribution mechanism of hardware offloading to avoid In the multi-tenant scenario, performance interference between multiple tenants is caused by the improvement of flexibility.
Multi-cluster performance level load balancing
The extreme IO performance flexibility introduces new challenges to the performance SLA, especially the IO burst performance limit of 100W IOPS introduces a greater risk of traffic congestion. For this reason, ESSD has designed a new multi-cluster performance level load balancing mechanism. The new intelligent balanced scheduling mechanism is composed of cluster/storage node/IO thread multi-level scheduling. According to cloud disk performance configuration, it monitors component IO load in real time, and realizes second-level IO load balancing in the cluster and minute-level inter-cluster traffic scheduling. When there is a significant performance level difference in the traffic between storage nodes, the cloud disk hot migration is triggered in real time, which solves the performance contention problem of users' large-scale cloud disk load at the same time.
Summarize
ESSD AutoPL, as the main product of ESSD in the future, covers all the industries and customers currently faced by flexible computing. The flexibility and elasticity of AutoPL reduces the difficulty of IT scale planning and the risks caused by improper planning. It will be favored by operation and maintenance personnel or IT resource procurement personnel. Whether it is a new customer of Alibaba Cloud or an existing customer, ESSD AutoPL can be purchased as an alternative to ESSD PL1. AutoPL provides customers with an economical, simple and convenient experience for sudden business growth. We look forward to using AutoPL products extensively and giving us valuable feedback to help us do better. We will continue to improve ESSD's performance and service quality assurance capabilities through technological innovation, improve user experience, and provide customers with non-stop computing services.
Original work: Alibaba Cloud Storage Xi Jian
Copyright Notice: content of this article is contributed spontaneously by Alibaba Cloud real-name registered users. The copyright belongs to the original author. The Alibaba Cloud Developer Community does not own its copyright and does not assume corresponding legal responsibilities. For specific rules, please refer to the "Alibaba Cloud Developer Community User Service Agreement" and the "Alibaba Cloud Developer Community Intellectual Property Protection Guidelines". If you find suspected plagiarism in this community, fill in the infringement complaint form to report it. Once verified, the community will immediately delete the suspected infringing content.
**粗体** _斜体_ [链接](http://example.com) `代码` - 列表 > 引用。你还可以使用@来通知其他用户。