(1) Accelerate the construction of container images, and pull multi-architecture large images/overseas sources at a high speed
The first stop of cloud nativeization is actually the transformation of business containerization. We need to package the application and the environment that the application depends on into a standard container image. In the process of container image construction, it is likely to rely on overseas sources (golang/yum/npm) or overseas basic images, and stable overseas acceleration is required. Or the business requires version output of multiple operating systems/multi-architectures, and concurrent matrix construction is required to test multiple versions conveniently.
ACR EE provides serverless container image construction services, supporting multi-task stable & high-concurrency construction. Support self-built GitLab/Codeup/Gitee/Github multi-code source binding, and automatic construction after code changes. Through overseas source intelligent acceleration and multi-level cache acceleration solutions, ACREE construction can be accelerated by about 30% on average compared to open source construction solutions. In addition, ACR EE also provides the ability to customize the version, build parameters, and the industry's first multi-architecture image build. After the support is completed, the experience upgrades such as trigger event callbacks are convenient for customers to more flexibly implement cloud-native DevOps scenarios.
The new version of ACR EE provides enterprise-level construction services, with features such as "multi-code source access", "large-scale concurrency", "multi-dimensional construction acceleration", and "experience upgrade".
(2) Accelerated global distribution and efficient synchronization of cross-border mirroring
Next, let us look at the acceleration of global distribution. At present, we have many Internet companies going overseas and multinational collaboration customers. Their R&D centers may be domestic, but the container business requires global multi-regional deployment and collaboration. Due to the instability of the cross-sea link, the problems of difficult coordination and slow iteration are caused. However, building a dedicated line link to maintain a complex network topology will bring huge resources and manpower operation and maintenance costs. In the end, these customers chose to use ACR EE's global acceleration capabilities to increase distribution efficiency by 85%, while reducing self-construction and operation and maintenance costs by 300%. The use of ACR EE's global unified domain name function enables intelligent access to nearby instances, further reduces the cost of customers using multiple instances, and improves service availability.
For scenarios where global distribution is accelerated, we have made multi-dimensional optimizations to improve and guarantee a 99.95% cross-sea synchronization success rate and low synchronization latency. We realize the optimization of the link layer through dedicated links, transmission protocol tuning, and multi-link QoS. Data slicing, data deduplication, and hierarchical concurrency realize the optimization of the scheduling layer, and ultimately ensure the ultimate customer experience.
The new version of ACR EE has enhanced "link optimization", "scheduling optimization", and "security compliance", which is especially suitable for companies with business needs to go overseas to achieve cross-sea CI/CD with peace of mind.
(3) The distribution of large images is accelerated, and the startup time of GB-level images is reduced by 60%
In fact, in addition to global distribution in many regions, more and more customers are also facing the problem of slow distribution of large images.
For example, after many AI services are containerized, he needs to package the algorithm model and algorithm dependency library in the container image for offline task processing. The single image may exceed 5G. As you can see on the left, when the traditional container is running, because the image layer is in the tar.gz format, it cannot be read randomly. Our Node node needs to download the full amount of container image data and then decompress it. However, the actual startup of the container may only use part of the content, which leads to the problem that the container startup takes a long time or even fails to start due to timeout.
ACR EE provides on-demand loading capabilities, which can automatically convert standard images into accelerated images. The accelerated mirroring layer is a custom indexed format that supports high-performance random reading. Realize on-demand loading and online decompression of mirrored data based on the acceleration kit on the Node node. The average startup time is reduced by 60%, the on-demand distribution efficiency is increased by 3 times, and you can enjoy the experience of running large mirrors in seconds. In addition, we also provide the ability to automatically switch to the accelerated image version based on the label, and support the use of transparent container services on ACK.
The new version of ACR EE is extremely friendly to large mirrors, with "on-demand loading", "accelerated mirror format conversion", and "experience upgrade", which is very suitable for AI/offline task applications, scripting language applications, and GB-level mirroring.
(4) Accelerated large-scale distribution, rapid deployment of a single cluster with hundreds of nodes
With the popularization of containerization in some AI/big data scenarios, the number of nodes in a single cluster is also increasing rapidly. For example, in some batch computing and genetic computing services, the number of nodes in a single cluster may exceed 100. When large-scale container nodes download images in batches, the network bandwidth of the image back-end storage (OSS) may become a performance bottleneck, resulting in slow or time-out image pull, affecting emergency expansion/change of the business.
ACR EE 提供了 P2P分发加速功能，利用计算节点的内网带宽资源，在节点之间分发镜像分片，大幅提升镜像拉取速度，减少应用部署时间。目前，ACR EE 支持单集群千节点的 P2P 分发，相比于传统的镜像拉取，分发性能提升4倍左右。ACREE也支持 ACK、ASK及 ACK+VK（ECS+ECI并存）集群模式下的 P2P分发，支持 Docker和 Containerd 的容器运行时。此外，ACREE 也支持基于 label自动切换至 P2P分发加速，比如在某个应用负载、或某个空间打上 P2P 标签，ACR EE 会将当前普通镜像拉取自动替换为 P2P加速链路，让客户的使用体验更加平滑。
新版 ACR EE 对大规模业务更加友好，尤其提升了100个以上节点的单集群的部署体验，实验数据显示对于单集群1000节点分发性能较此前提升4倍。Based on the container image service enterprise version ACR EE, from container image construction, global distribution, large image/large-scale distribution, accelerated upgrade of the entire link, to improve the efficiency of cloud-native application construction, distribution, and deployment, and help enterprises to upgrade their cloud-native architecture.