Shulou(Shulou.com)05/31 Report--

This article introduces how to carry out the theoretical analysis of Kubernetes as the Framework of Mesos. The content is very detailed. Interested friends can use it for reference. I hope it will be helpful to you.

When looking for a container orchestration platform, the open source community offers some viable options such as Kubernetes, Marathon-Mesos, or Docker Swarm. Among many users looking to run an online load native cloud, Kubernetes is very popular because it has built-in support for a range of useful features, including automated deployment, load balancing, automatic extension, rolling upgrades, and other elements that DevOps users are looking for. But at the same time, considering the load scope of the enterprise operating environment and dealing with issues of common concern among multiple frameworks, a more general resource manager is needed, and Apache Mesos solves this problem. Mesos is a lightweight system with only 230K lines of code, while Kubernetes has 1.3MB, which focuses on providing a basic resource abstraction layer that allocates resources by framework, handles task-based execution capabilities, and handles administrative operations such as host maintenance mode.

At IBM, we explored how to make Mesos and Kubernetes work together by using Kubernetes as a Framework of Mesos. Although there is some overlap between Mesos and the resource management functions available in the built-in Kubernetes, the two tools can still be used for collaborative management. One of the advantages of this model is that it not only has the ability to support some Spark loads like big data, but also allows other analysis systems to benefit from the fine-grained resource allocation supported by Mesos. We believe that this will improve resource usage and performance when running highly dynamic business loads.

Because Mesos is lightweight, thousands of nodes can be easily dispatched in a production environment. On the other hand, although Kubernetes is rapidly improving its scale and performance, it is still limited to 1000 nodes. By running Kubernetes on Mesos, you can provide the ability of a Kubernetes node to run several times over Mesos.

Kubernetes only focuses on containers, while Mesos is more focused on supporting load at the operating system process level. It is more important for enterprises to put the existing business load into a jointly managed environment. Kubernetes allows enterprises to enter a new native cloud of containerized business loads, while Mesos is a bridge between existing Windows, Unix, and Linux environments.

As part of the exploration, IBM will be the first to contribute to the project to integrate Kubernetes into Mesos. This includes trying to help kube-DNS with external DNS (# 28453), improving the scheduling algorithm of Kubernetes (# 31068), making Kubernetes namespaces work better with Mesos (# 31069), and supporting Kubernetes to run on heterogeneous hardware environments (# 29901). This work is an additional contribution to IBM's ongoing direct implementation of Kubernetes and Mesos projects.

As to how to package these technologies in a consumable way so that ordinary enterprise users can easily build their own container management platform for different workloads, we will share more information soon, please stay tuned!

Source link: Exploring Kubernetes as a Mesos Framework: Does it make sense?

The translator said

Mesos is defined as a very large-scale cluster management platform for scheduling hundreds of nodes, while Kubernetes is defined as an open source system dedicated to managing, deploying, and scheduling containerized applications. After Google announced Kubernetes open source in June 2014, Google officially announced a partnership with Mesosphere in July of the same year to provide users with the ability to share Mesos cluster resources with other Framework (such as Hadoop, Spark, Marathon, Chronos, etc.). This collaboration can be understood as integrating Kubernetes into the Mesosphere ecosystem as a Framework. From a functional point of view, Mesos itself is more like a resource management tool of IaaS layer, which is used to provide the upper framework with the ability to share underlying resources, and then run Framework above to realize the application management of PaaS layer, which can schedule and manage different Framework, while Kubernetesm is just a Framework of a scheduling management container. The article says that Mesos has only 230K lines of code, while Kubernetes has 1.3m +. Mesos has less code than Kubernetes because the Mesos framework is incomplete and many other "Framework" (such as Marathon, Aurora, Singularity) are needed to schedule like Kubernetes. The two products have their own advantages and disadvantages, and the purpose of this cooperation is to provide a platform to meet more needs of users through the combination of the two.

Although this project has been proposed since Kubernetes opened source, more than two years later, it has not been effectively promoted in production practice, and now this project is dominated by the Mesos community. The code contributions from the Mesos community are mainly Mesospher, Twitter, IBM and Intel, while the Kubernetes on Mesos project is now mainly promoted by IBM, and Intel recently said that it will continue to invest money to support the Mesos project. They believe that although Kubernetes has made a great breakthrough, they do not think that Mesos and Kubernetes are competitors. Kubernetes is directly for application developers, while Mesos is for very large-scale cluster deployment, which is more suitable for data centers. As mentioned in the article, Kubernetes currently manages only 1000 nodes, and by running on Mesos, you can expand Kubernetes's capabilities several times to make a larger resource management platform. Intel believes that the goal of Mesos is the resource management and orchestration of large data centers, and the distinction will be drawn when running Kubernetes on Mesos-managed data centers. The following is a structure diagram of Kubernetes in Mesos:

Figure 1 Kubernetes on Meos Source GitHub

Kubernetes on Mesos does not simply run Kubernetes on the Mesos platform, but integrates Kubernetes into the Mesos project and solves the corresponding problems. As mentioned in the article, IBM is dealing with some of the issue, such as trying to help kube-DNS and external DNS (# 28453) and so on.

So does Kubernetes make sense as a Framework of Mesos? We can analyze the advantages and disadvantages of the combination of the two in production practice.

Advantages:

Mesos provides to run multiple Framework in the production environment. When the production environment not only needs to run container choreography Kubernetes, but also needs to run other Framework, this mode allows Kubernetes and other Framework to share resources, which is helpful to improve the resource utilization of the system.

Through this mode, Mesos can enjoy Kubernetes's abstraction of micro-services and implement a set of containers running on the same slave node.

At the same time, Mesos can automatically expand the Worker node of Kubernetes according to the load, without the need to install it manually.

Kubernetes can also be expanded several times without being limited by the number of nodes.

However, standing on the current time node, the shortcomings of this scheme are also very obvious:

The advantages mentioned above can only be realized in the face of super-large data centers (10W + physical nodes), but how many domestic enterprises can meet the super-large cluster scale? And domestic companies like BAT do not use it, they will develop directly on their own.

The open source ecosystem is unstable, the three kinds of COEs of Docker are still in development, the functions are gradually improved, and the product iteration is very fast. Moreover, there is a competitive relationship among the three kinds of COEs, and the result is not clear.

The technical solution is too complex, it will be more complex and heavier in production practice than using only one product, and there will be more holes. The two products themselves have overlapping functions. from a certain point of view, these two products are still similar products for users. If one can meet the needs of users, why use two to be so complicated at the same time. And ordinary enterprises will not make themselves so heavy, will be step by step to implement some technology, otherwise it will be indigestion. Small complexity and rapid implementation is the key for enterprises to choose new technologies. Too complex architecture can not be solved by general enterprises, which will only lead to a longer practice cycle, will affect long-term stability, and landing will become more difficult. Although the function of this model is more comprehensive, it also increases the difficulty of landing, and some enterprises do not need this function at this stage, so it will be appropriate to choose a single product.

Mesos is lightweight and has less code than Kubernetes, but the combination of the two will increase the weight of each other, which is a contradictory setting.

There is also some functional overlap between the architecture of Mesos and the popular Openstack positioning, so that enterprises will not know how to choose the landing solution, so the choice of this solution will be very low.

This is the end of the theoretical analysis on how to conduct Kubernetes as the Framework of Mesos. I hope the above content can be of some help to you and can learn more knowledge. If you think the article is good, you can share it for more people to see.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.