Alibaba Cloud Unveils ECS Core Technologies at LC3 Conference 2018

On June 25, 2018, the Open Source Summit LinuxCon + ContainerCon + CloudOpen (LC3) hosted by LF Asia, LLC took place at China National Convention Center in Beijing. The conference brought together thousands of professionals from around the world including developers, architects, system administrators, DevOps experts, and business leaders.

During his exchanges with many big names in the industry at LC3, Chu Ba, an Alibaba Cloud researcher, spoke of the challenges that Alibaba has met so far in promoting the ubiquitous usage of computing capability and cloud migration, which, on the other hand, have greatly enhanced Alibaba’s technical capabilities and accumulated experience.

Two Key Accomplishments by Alibaba Cloud ECS in 2018

At the much-anticipated Apsara session, Alibaba Cloud ECS team unveiled the ECS core technologies by sharing topics such as performance improvement for artificial intelligence (AI) applications, live migration, and qcache performance improvement in virtio-blk. The team also shed new light on the improvement of virtual machine performance.

Senior technical expert Yun Long talked about the optimum acceleration of AI services. He put forward several solutions to AI scenarios based on multiple application optimization instances. For example, in a scenario of click-through rate (CTR) prediction training, he demonstrated that application performance could be improved by 13 times through optimization of GPU distributed training.

Technical expert Zhang Chao introduced live migration, one of the technologies that ensure the high reliability of Alibaba Cloud ECS. The live migration technology allows a running virtual machine (VM) to migrate from one physical host to another with no downtime and no influence on user experience.

Senior engineer Nian Qing expounded on the qcache performance improvement in virtio-blk in various aspects, including KVM storage stack, qcache operating principle, common qcache usage modes, qcache performance, and the future prospects of qcache. After introducing the basic operating principle of qcache, Nian discussed the two features of qcache: processing of mass missed requests and seq-cut-off. When there are mass missed requests, qcache uses the PT processing mode to relieve the stress from the cache and maintain an appropriate speed of response to guest requests. The seq-cut-off feature allows the system to auto-determine whether to cache a large chunk of continuous requests. At the end of the sharing, Nian gave her insights into the development trend of qcache, including the support for spdk, optane, and QOS, and the dynamic allocation of the cache size as well as qcache open source.

Senior technical expert Zhe Yang analyzed the performance issues of typical business scenarios in the virtual environment, and provided a new technical solution, which can effectively reduce the virtualization expenses and enhance the VM performance through a few modifications to the kernel. In practical tests, this solution can improve the system performance by about 20%.

To learn more about Alibaba Cloud Elastic Compute Service, visit


Follow me to keep abreast with the latest technology news, industry insights, and developer trends.

Follow me to keep abreast with the latest technology news, industry insights, and developer trends.