Shulou(Shulou.com)06/02 Report--

Nowadays, Internet enterprises rely on technological advantages, profoundly affecting and changing people's lives and work, among which open source technology has given birth to the development of Internet enterprises. Behind cloud computing, big data, AI and IoT is the support of open source technologies such as OpenStack, container and Hadoop. Under the support of open source technology, Internet enterprises are like a fish in water.

Compared with Internet enterprises, traditional industries / enterprise users mostly adopt the mode of IT service outsourcing, which mainly depends on the technical capabilities of IT product suppliers, ISV and SI. The technical strength and talent accumulation of the industry itself have no way to compare with Internet enterprises. Under such circumstances, how should enterprises in traditional industries deal with open source technology?

In this regard, Geng Hang, deputy secretary-general of the China Open Source Cloud Alliance and co-founder of the Ceph Chinese community, gave his views in the keynote speech of the 2018 China data and Storage Summit.

Geng Hang said that what suits you is the best choice. Industry enterprise users can choose their own cloud platform, storage and network according to their own needs. The characteristics of open source products are different, and the suitable application scenarios are also different. Take OpenStack as an example, its characteristic is comprehensive, compared with CloudStack, ZStack is relatively lightweight, easy to manage and use, therefore, the characteristics are different, the user's choice is not used.

For industry enterprise users, in order to adapt to the characteristics and changes of the cloud computing era, the user team needs to do the following things:

In Geng Hang's view, this is a feasible adjustment and change.

In addition, combined with industry practice, Geng Hang also shared his experience on network problems, disk problems, time issues, resource acquisition problems, and application scenarios in Ceph deployment.

The following is the full text of the speech:

Geng Hang: good afternoon, everyone. I am the last one to speak today. Thank you very much for holding on to the end. To introduce myself, my name is Geng Hang. I currently work for China Railway Trust. My social position is the Deputy Secretary-General of China Open Source Cloud Alliance and the co-founder of Ceph China Community. Today, I will mainly talk about the suggestions and thoughts on the landing of open source cloud technology in traditional enterprises.

First of all, it is divided into these aspects: first, we will introduce the China Open Source Cloud Alliance and the Ceph Foundation. Later, we will introduce some open source ecological development in China, including some thoughts on open source cloud technology, and finally talk about how traditional operations and maintenance can learn quickly, such as Ceph or some related cloud technologies.

The China Open Source Cloud Alliance, currently affiliated with the China Institute of Electronic Technology Standardization under the Ministry of Industry and Information Technology, is mainly engaged in domestic cloud, standards, and open source related industries. Domestic cloud enterprises are now in it, including doing some open source standards, or having exchanges with international foundations such as CNCF.

The Ceph Foundation, if you know Ceph, Ceph just established the Ceph Foundation last month this year. After the establishment of the Ceph Foundation, it is a very good development for Ceph, because it has been more than a decade since Ceph was founded in 2004, but it has been acquired by Red Hat in 2014. After the acquisition, Ceph's software and community are owned by Red Hat. We all know that Ceph is very popular, and many ecological manufacturers are involved. In this way, there is a bottleneck in the future development of Ceph, and we will not make use of its future construction. Under the promotion of the founder of Ceph, the Ceph community set up the Ceph Foundation in November this year. The Ceph Foundation is affiliated with the Linx Foundation and is at the same level as the CNCF Foundation. Sponsors who participate in the foundation will have some rights and interests, such as participating in the decision-making of its activities, including some of its future technological development, and so on.

The following introduces some users in China, the top layer is its top users, China Mobile, including ZTE, XSKY. The second is an enterprise member, including Didi, EasyStack, Ping an of China and QCT. The third layer is SRIE Tiexin, which is specially invited by users and research institutions, and also participates in the foundation.

If you know Ceph, or if you know OpenStack open source cloud technology, you can get videos and some information by following our official account (cephchina) and knowing that we have held a conference for the year of Cephalocon APAC2018. This has also been operated for a long time before landing in China. At present, in terms of open source cloud technology, including the field of OpenStack open source cloud technology, none of them said that the first international summit was held in China, so we also played a great role in promoting domestic open source cloud technology. We have been negotiating with the international community that the first international summit must fall in Beijing, and there are many users and developers in China. In the end, people with good intentions landed in Beijing, and then a lot of foreigners came. The next Summit will be held in Barcelona, co-hosted with KubeCon of CNCF.

After talking about the background and introduction, let's talk about the development trend of domestic open source culture.

First of all, open source is not equal to free thinking change, like a few years ago, when it comes to open source, the first thing that comes to mind is free, but in fact, open source is not free, just like farmers' springs, we do not produce water, but we are the porters of nature. it's the same reason. Up to now, we have gradually accepted the concept and thinking of open source services. Just now, the money of Commvault always talked about the change of users' thinking, and now we begin to accept it gradually.

There is also a change in the thinking that open source is unsafe. In the past, when it was mentioned whether open source was safe or not, there would be such considerations. It can be said that there is no absolute security in this world, only relative security. The open source project is a good example of relative security. At least it is open code that anyone can read and change, so its security is better than that of closed source. Here I list some domestic users, including some operators, the Internet, the financial industry, the game industry. In fact, these users are also using open source cloud software such as Ceph or OpenStack.

There is also an open mind that embraces open source. In the past, everyone was doing open source at home, making a small fuss about an open source software at home, and then promoting it at home, all playing at home. Now we can see that some open source projects such as ApacheKylin, Habor, Tidb and other open source software have been gradually promoted to international open source foundations. For example, Habor has joined the CNCF Foundation, ApacheKylin and SS have also joined the Apache Foundation, and all of them have joined the International Open Source Foundation accordingly. Open source culture has changed from embracing to opening up, from domestic to international.

Next, let's talk about the landing thinking and suggestions of open source cloud technology in traditional enterprises, as well as some viewpoints. Here we also refer to some views of many friends and enterprises in the circle, such as my good friend Liu Shimin. The director of cloud research and development of an enterprise has personally experienced the experience of cloud transformation in the enterprise, and he is also a technology network celebrity in the circle.

OK, back to the point, everyone mentioned earlier that enterprises will go to the cloud, so the cloud will be the norm in the future, including the big strategy proposed in the future, and the edge of the cloud tube & the edge of cloud data. Everything will be based on the cloud in the future. The cloud in the enterprise is the normal state of the cloud. I'm talking about its infrastructure, the cloud on the infrastructure and the cloud of the business. The cloud on the infrastructure is very simple, the hardware server, storage on the cloud. In business, the cloud is the migration of stand-alone business applications that used to be local to the cloud.

What I'm talking about here is the selection of its technical software platform, that is to say, what is suitable for you is the best, and some are blind selections. Now the most popular cloud platform is OpenStack, but you don't necessarily choose OpenStack, what is suitable for you is the best. Of course, many people say that the OpenStack architecture is very heavy, how to say this thing, open source products are large and comprehensive things, it can not be said that some functions it does not have, but can be used is the same thing, open source products and the actual landing is not the same, if open source products can be directly used, everyone does not do products, directly do services. Suppose it has 10 components, not all 10 components, according to the actual application scenario, just say private cloud, use its core components as private cloud. The advantage of CloudStack and Zstack is simple and easy to use, and its components are much less than OpenStack, which is also related to the ecology of OpenStack. Why OpenStack is heavy, everything is included, edge computing includes everything, and it also has something to do with its operating ecology. More vendors join in order to invigorate open source projects, and there are so many things that you may not use in practical application.

Including the selection of storage, the most popular is Ceph, of course, SheepDog, GlusterFS and so on. In many cases, you don't have to use Ceph for storage. Of course, there are many benefits of using Ceph. There are many ecological aspects of Ceph, from the underlying hardware Intel and ARM, including the upper operating systems Red Hat and SUSE, to the upper cloud platforms OpenStack, CloudStack, and Zstack, which all support Ceph, including the top application OwnCloud. So we can see that its applications from the bottom to the upper layer are supporting Ceph,Ceph. It is a very perfect open source ecology and community ecology. In fact, when choosing open source software, it is nothing more than these points, first of all, choose open source ecology, choose open source ecology is nothing more than save manpower, time and other costs.

In addition, there are network selection, such as SDN, VXLAN and VLAN. In fact, I don't think it is necessary to use VXLAN in private cloud, which increases the cost of operation and maintenance. VLAN itself satisfies the scene of private cloud, and it is more difficult to use VXLAN. It is nothing more than digging a hole for yourself to jump in. If something goes wrong, it is not easy to troubleshoot.

Like when I was doing public cloud network research open source software before, we selected a company whose software is very good. It can be perfectly combined with OpenStack to replace OVS in OpenStack, including its advanced features, such as × × AAS, FWAAS and other functions. This thing is very good, but it does not have a good open source ecology. If the company dies, the project will die with it. No matter whether we choose open source at home or abroad, we still have to consider the open source ecology very much, including why SheepDog and GlusterFS are seldom used because there is no good ecology.

Going to the cloud is only the starting point rather than the end, because it involves technical factors, team factors, cooperation among multiple teams, and cooperation among many teams in traditional enterprises, such as server operation and maintenance, database operation and maintenance, and so on. Including some migration, just now also talked about data migration, including virtual machine migration, V2V and so on, there are many factors in it.

You can see that many technical terms have been written on my PPT. In fact, in the early days of Shangyun, it can also be seen as a transitional period. Why do you say so? the early stage must be the state of coexistence of open source and commerce. As EasyStack said just now, their storage is the coexistence of commercial storage and distributed storage, gradually separated from commercial products to open source technology products. Why do you say it's just the beginning, not the end? Because you have a lot of things to do, as mentioned by EasyStack just now, they have moved a lot of open source technologies to traditional enterprises, but it is also the same. For example, technological concepts such as DevOps and CI/CD are very popular, but it is very difficult to land in traditional enterprises. DevOps and CI/CD are very good, and the ideas are also good, but they are a little different from the reality. This thing is a complete innovation, subverting the habits of previous development. It is even said that your previous habit of thinking may have been operated locally, but now it is different and you should concentrate on it. Therefore, it is completely a kind of subversion, and it is very difficult to promote in traditional enterprises. There are only a few factors to promote. First of all, this thing must be top-down, including the top-down cloud of the enterprise. If it is the bottom to promote is actually very difficult, of course, it is not 100% success, sometimes there are halfway from top to bottom.

Like some of the transformations mentioned earlier, including some big strategies (cloud data edge). In fact, it is the same on the railway now, based on the cloud, using big data, edge computing IOT and other technologies to better improve some railway systems, including saying that the railway is now building a second-level management mode. In fact, in the past, it was the general railway, the railway bureau and the railway section, but now it is gradually changing from the third level to the second level, gradually controlling some things, data and resources. In fact, some of the technologies mentioned just now are also used, including the cruise of drones, the analysis of big data, and so on, to better protect the railway system.

There are many enterprises on the road of digital transformation, not only the railway, but also the manufacturing industry. They know where the direction is under the country's general policy and policy, but they do not know how to do it. As far as I know, there is a state-owned enterprise of military supplies in Shijiazhuang. They are also seeking digital transformation. When they seek digital transformation, they do not understand it, but they do not know how to do it. At this time, they lack some guidance. Including some advice on consultation and landing.

As mentioned just now, there are only a few main points: first of all, there must be a suitable cloud service provider. Why do you say that? because the service provider needs to understand the business very much. Only by knowing your business and your actual situation can we make a corresponding cloud plan. Second, Shangyun needs the cooperation of multiple relevant technical teams, not only a single operation and maintenance team, but also multiple operation and maintenance teams to cooperate with each other. Third, at the same time, it is necessary to evaluate Shangyun's plan and the development of migration tools. Fourth, we need to advance in a safe and orderly manner, step by step, and cannot be achieved overnight. Fifth, we need to fully consider the issue of security, many of which consider the issue of security after going to the cloud. As mentioned just now, infrastructure on the cloud, server storage, including secure devices are also on the cloud, now there are a lot of security on the cloud and so on. Sixth, we should also consider technology, cost, team development, business issues and so on. Seventh, team training is mainly about team training. in the early days of cloud in traditional enterprises, it must be to find a cloud service provider to help me build my business quickly. while building up, we should also build our own relevant cloud team, in which we should run-in with each other, learn from each other, and train our own people at the same time. When it comes to the middle stage of training, the middle stage certainly means that one's own people will take over part of the work of operation and maintenance and continue to train troops. In the later stage, the team has almost run-in, and can have some relevant cloud development and work, that is, they can do what they can do on their own, and then use external service providers when they cannot do it.

Finally, uh, let's talk about some problems encountered by traditional enterprise operations and maintenance during the transformation of Ceph. Beginners will encounter network problems when playing Ceph, because the Ceph source is abroad, and reading official documents is also using foreign sources. Foreign sources are easy to time out. What to do, coupled with domestic sources, many domestic sources, such as Ali, NetEase, China University of Science and Technology, and so on, have many domestic sources. Another problem is the problem of the disk. In fact, I think it is the last thing to say about the disk, because if you want to play with a storage or something, you must have a clean disk, including the problem with the latter disk, that is to say, in the new version, because it was possible to do its OSD based on the directory in the past, but this is not allowed after the new L version. Because this mistake is not allowed to do so, may have played before, and then did not play for a long time, play the new version, there is this problem. How to solve this problem? You can use DD method, and then LOOP way to hang it locally, but it is not recommended to do so. This can solve the problem, but it is not recommended, because the purpose of doing this is to test, test to select virtual machine, physical machine, assuming that you do not lack a disk, and this problem will not occur if you do not lack disk, so it is said that some problems are actually caused by itself in the process of self-study. And then accidentally fell in.

In other words, why does another problem arise? because three nodes are officially recommended, but he has to deploy with a single node, there will also be strange problems of one kind or another, so you must follow the official way, not according to your own understanding.

There is also the question of time. In the following figure, we can see that the process is still there, but the status is down, troubleshooting for half a day is a matter of time, and the state is inconsistent due to inconsistent time nodes. Another situation is the case of the network card, which has been encountered before, the OSD status is now down and then up, and this situation will also occur when there is a problem with the network card. In fact, there is also the question of time, for example, Ceph has a component called, Monitor,Monitor is very time-sensitive, once more than how many seconds this state is not healthy.

Having said so much before, I have also seen from the summary of some questions that this is the first problem, that is, its domestic mirror source, and the domestic documents, including community groups, can be set up, and you can consult it about it. For example, OpenStack and K8S are embracing Ceph. How to say this? many people say that Ceph is not well supported, but in fact, it cannot be said that Ceph does not support K8S well, K8S does not support all blocks, and does not support all blocks, only for distributed file storage. Now some of the domestic players in K8S are RBD,Ceph blocks of Ceph and useful CephFS. This piece is still not recommended, because there are still a lot of pits, and there are many pits when there is a large amount of data. I forgot to mention the selection just now. There are also some selections, that is, when you choose the type blindly and do not fully consider the actual business situation, you blindly select the storage software. For example, CephFS, many domestic customers can use object storage to solve problems, but it uses FS in its selection. When it encounters problems after a certain amount of business, it is forced to transform to the object, and the data is moved to the object. There is also a company before the wrong selection, the use of FS, and finally improper operation of the data lost, these situations are actually some of the problems caused by the previous selection.

The next step is how to use it after deployment. The most typical scenario is the block application, which just docked like ZStack, which is very similar to Cloud Stack. Those who have played Cloud Stack think ZStack is very interesting. And OpenStack, all the mainstream versions of Ceph are compatible, why every time the Ceph Foundation publishes a survey report, Ceph always ranks first, this is a very important point, its compatibility is very good, the compatibility for the OpenStack community is very good, why is it more advanced than commercial storage, because commercial storage vendors do driver version development for certain versions, and some users will be incompatible after upgrading, and so on. That's why Ceph ranks first on OpenStack.

Then there is the public cloud based on OpenStack, and even the private cloud has the confusion of capacity and performance disk. When you create a disk, it will let you choose between capacity disk or performance disk. How to achieve this? based on Ceph, it is very simple, because Ceph has an algorithm called Crush algorithm, which has a CrushMap that can be edited to compose the disk logic on my actual physical machine into a host, or multiple host, to form a fault domain. Suppose you classify all SSD as host, associate several host with SSD, SATA and Pool, and associate Pool with Cinder of OpenStack. Cinder can support multiple Type, support multiple backends, and correspond to the type of multiple disks. This is the same. The implementation through Ceph is implemented with CrushMap editing. The concept of Ceph failure domain can divide whether your data is distributed by disk or by machine or rack, and you can adjust how you want to divide it. At the same time, in the concept of failure domain, you down a machine in the fault domain, and the rest is not affected.

I have a native page of Ceph here. Ceph came out after the L version. The L version is a milestone version of Ceph. There is no good page before the L version. After the L version, there is a very perfect product page. After the L version, there is more and more production. You can see that it is very perfect. This is the main page. This is some of its storage capacity, including some of its storage Pools. Some of its logs are very production-oriented. Now all open source distributed storage does not have such a perfect page, including some nodes of the cluster that can see it, including resource pools, including the blocks that created it, including deleting blocks, including snapshots of blocks, deletion of snapshots, snapshot locks, including its objects. Ceph is unified distributed storage. Provide both blocks and files and objects, its object is to create Bucket operations, create users to delete users, including AK/SK can be seen on the page, and command-line operations are liberated to operate on the page, gradually moving towards production and simplicity. I'm not going to play it here. The main functions are the functions I just mentioned.

Then unstructured data is everywhere. In fact, we all know that some live broadcasts on mobile devices are very popular. After live broadcasting, it has playback function. Playback function is unstructured data. Where it is stored, it is stored in object storage. Domestic Huya TV stores the video on the Ceph object, including Vphotos, which is actually cloud photography. The organizers who often attend the meeting know that where you can see today's photos, the organizers say you can see today's photos. After taking photos, there will be such a logo in the lower right corner. In fact, they are all photos and unstructured data everywhere.

The other is the application of object storage, Ceph object storage. In fact, Ceph object storage is currently the most widely used, that is, you seem to be playing games and backing up documents, including some time ago in NetEYun, NetEase almost all products, such as koalas, are used in object storage. For example, Ctrip is also using object storage, including Jinri Toutiao, some of its videos are stored on objects, including China Telecom, where CDN is docked with Ceph. The scene of object docking is streaming media, transcoding, network disk, including some network disk. The simplest thing is that Own cloud native is supported by me just now.

After that, it may be said that since storage must consider its disaster preparedness, it was also mentioned just now that Ceph can do some failure domains based on cabinets. After the downmachine, the business in other places is not aware, and it has no impact on the business. In addition, there are some disaster recovery of blocks, which can be replicated across clusters. There is object-based multi-site disaster recovery, can be multiple sites for disaster recovery, some people may say, you only talked about blocks and objects, not files. As mentioned just now, although the document said that it could be used for production, it was not recommended before the L version, and it was not recommended to produce before the L version, but it is no longer recommended for production after the L version, but it is still not recommended for ecology, because there will be more IO generated when there is a large amount of data, including when you migrate, it actually has a lot of problems, and it does not say that it meets some of the characteristics of FS. So it's not recommended yet, so I don't recommend how to use FS.

In fact, there are many scenarios, such as objects, which can completely replace files, including, for example, the official Ceph community converts the interface into NFS on object storage, and you can use it. For business applications, it is a file, and the object is stored behind it, and you can also play it the same way. Including the latest ones in the community, those mentioned earlier in the hybrid cloud can also transfer your local data to the public cloud object storage through the latest Cloud Sync after you deploy the object storage locally. Of course, you can also pull over their data objects, which are also data management and backup, and can be backed up to the above data.

After talking about installation, use and disaster preparedness, it is now operation and maintenance. I recommend one here, that is, operation and maintenance on the command line, including some people say that tools that encapsulate API, including some commands, can be used for secondary development. In terms of monitoring, many people do not know how to monitor this thing after use. I said that the native Dashboard I just mentioned has some microscopic monitoring, and you can see the read and write status of the OSG. And Zabbix, which has a Ceph plug-in. The other is to play with Prometheus, which is now mostly used to monitor Ceph. Two management classes are recommended, and these two can also be deployed. If you play with Ceph, you can be regarded as a conscientious open source product. The previous PetaSAN focuses on blocks on the native Ceph, does some work on the native Ceph, and does some work on the blocks, including manageability, weakening monitoring, and deployment. The latter Proxmox is similar to ESXI virtualization, and can also be managed to add and delete. These two products are open source, and if you look at its interface, they are better than the original ones. These two products are open source products with a conscience of the industry, and they still follow closely with the community version.

Later recommend a book, rookie words can follow our book "Ceph distributed Storage practice", I am also one of the editors and authors of this book, this book is to help Ceph rookies and outsiders quickly understand what Ceph is, what Ceph can be used for, and even the application scenarios and basic operation and maintenance of Ceph.

There are also some ways to learn. I just mentioned some learning ways for talent training, in addition to running-in with Party B's cloud provider team, but also through peripheral websites and online schools. Through online school learning + Party B cloud providers run-in with each other, can quickly and effectively help their own team building effectively.

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.