In the era of cloudification, live migration of virtual machines is a hot technology that has attracted widespread attention. As one of the core technologies of cloud computing platforms, live migration of virtual machines is also the underlying technology that Convince Cloud has been deeply researching.
So what is the strength of the hyper-converged products under Convinced Cloud in terms of hot migration? This test revealed the answer.
Virtual machine live migration refers to the migration of virtual machines from one host to another to ensure that the business always runs on a physical host that is reliable and has sufficient resources, and the business is not interrupted as much as possible during the migration process.
As more and more users deploy core services on the cloud platform, business applications require higher SLA guarantees, and the requirements for the window time for downtime are getting higher and higher, and even zero downtime is required. Especially in the following scenarios:
(1) Maintenance of physical hosts, such as parts replacement and service maintenance.
(2) Cloud platform upgrade requires rolling hot upgrade of physical hosts in the cluster.
(3) The cloud platform schedules the running location of virtual machines in the cluster to obtain better resource utilization efficiency.
In order to test the real performance of the virtual machine hot migration of Trust Cloud, Trust Cloud tested the live migration of virtual machines under the Oracle service in a real environment, and compared it with other manufacturers in the industry.
test environment
Test Results
↑ Sangfor Hyper-converged Live Migration in Oracle 1000 User Scenario
Result description
From the above test results, it can be seen that the hot migration solution based on Sangfor’s hyper-convergence can achieve smooth business continuity experience, and the migration is faster and more stable.
For the Oracle service virtual machine carrying 1000 concurrent users, the performance of the Oracle service hosted by Sangfor hyperconverged only jitters for 6s during the migration process, and the service interruption time is only 1s, and the upper-layer service can get a smooth migration experience with almost zero interruption. Compared with the test results of other hyper-converged products in the industry, Convinced Cloud has obvious advantages.
Innovative design behind smooth migration
Sangfor hyperconvergence can achieve a smooth migration experience far superior to other products in the industry, derived from the advantages of multiple underlying core technologies. One of the key technologies is the CPU throttling algorithm, the basic principle of which is as follows:
In the initial stage of migration, the source host starts an identical virtual machine on the destination host through the negotiation mechanism and puts it into a quiesced state.
Then, after the source host QEMU records all the memory data of the virtual machine, it is transmitted to the destination host through the network. However, since the running business will read and write memory data during the transmission process, the newly generated memory data will be re-recorded after the first round of transmission is completed. And transmit it to the destination host in the same way, and iterate in turn until the last recorded memory data meets a threshold, the source host will stop running (that is, only 1s of service interruption duration).
QEMU transfers the remaining memory data to the destination host at one time, then the destination host starts running, and the source host shuts down. In this process, it may happen that the sequential iterations fail to converge, which will reduce the CPU utilization of the virtual machine (that is, throttling), and reduce the throughput of the business. The optimization of Sangfor hyper-convergence is mainly by designing a new algorithm, taking into account multiple indicators. , which takes into account the migration time and the time that the business is affected by throttling, which can ultimately reduce the migration time and the time that the business is affected by throttling.
The implementation of this technology enables the scenarios that could not be migrated to be successfully migrated now, leading the industry. At the same time, during the final downtime of the live migration, the ping network interruption time does not exceed 2 seconds (the best test result is 0s). .
In addition to the CPU throttling algorithm, Sangfor hyperconvergence has optimized the QEMU hot migration compression algorithm, the disk hot migration BITMAP data block granularity adjustment, the hotspot memory dirty data delay migration mechanism, the memory hot migration time slice dynamic adjustment mechanism, and the disk hot migration IO QoS dynamic adjustment mechanism. Mechanism and other technologies have their own innovative designs, and the follow-up “Convinced Cloud Blackboard News” will bring you the sharing of Convinced cloud technology dry goods.
The above is the sharing of virtual machine hot migration in this issue of “Convinced Cloud Blackboard News”. Pay attention to the “Sangfor Technology” official account to get more technical dry goods content!
Leifeng Network
This article is reprinted from: https://www.leiphone.com/category/industrynews/UY4M0cU7RWs41D1X.html
This site is for inclusion only, and the copyright belongs to the original author.