Hi everyone, I'm Xiaowu Han. I'm the director of Ecosystem.Hello, I'm Xiaowu Han.歡迎大家來參加我們本場的小部討論,也就是利用原生技術來進行企業轉型中國地區,首先我會讓我們講者來進行一個簡單的做介紹,大家可以用中文也可以用英文,首先我是Xiao Heng.我是Lin Zhang,來自...大家好,我是方來自法威爾.我主要是在執行法院,我主要是在開放技術,我在執行法院,我主要是在執行法院,我是Xiao Heng,來自Tihu 360,我主要是在執行法院,首先第一個問題就是,能否來介紹一下你現在正在工作的這個原生的項目呢?我會介紹在中國,其實我們的專頁程式,其實是經常使用的,無論是 container 還是 non-container,也許是Cubinators, container, etc.也許是non-container,也許是用了etcd,也許是用了這些,也許是用了一些 non-container,也許是用了一些 non-container,也許是用了一些 non-container,因為大量的儲物,我們改改了一些非常大的版本,We have some modification or some development and this lead to that in the old version there will be big changes and difference.So first maybe we are not able to update it.And the second when we are using these software some applications they cannot update themselves also.And so our solution is to fork a complete version into our company so that we can use the version in our company to update it and to upgrade it.But we should not change the source code actually because it is of a large sum of work.And future in its scalability actually performs quite well.So yeah this is the brief introduction.Well I will also briefly introduce it.Actually in Alibaba I specialize in container and also we begin in previous years.And in previous years we already finished the transformation of the program in Alibaba and we have accumulated some experience for example Sigma.It is also container based.It is a scheduling or orchestrator.Well in recent years we are trying to make the system more collaborative and to migrate it.And during the migration we also encounter some challenges.Maybe the biggest challenge is the scalability or the performance of the Kubernetes.But when we are actually transforming it.Maybe mainly it is about the matching of the upstream and the ecosystem like the PERS.Also the developers in their processes they may encounter some challenges.So without changing the cloud native system we need to cope with these challenges.And the second is how to maintain about its development and maintenance.And its scalability.Even if you do not violate the source code you can also use it.But the problem is how to maintain it especially in Kubernetes.And in Alibaba we have thousands of hundreds of nodes.So how to control it without impacting on our business.So in development and maintenance we encounter a lot of challenges.And in our company mainly while we are since 2013 we began working on the PERS project.And now we also have part of the cloud native technologies.Well actually its earliest version at that time.Inspiration has high accomplishment.But overall it has low customization.So since 2014 we migrated to Kubernetes.And during this process until now we have served the private cloud.A large number of clients and the public clouds as well as public clouds.So in early days we encounter some problems for example for some traditional use.Traditional customers they have special requirement for the distribution.The allocation for example they should be distributed on different nodes.And in terms of the network they hope that the traditional design can be separated or be spliton different control planes according to different requirements.At that time what we mainly worked on is on the upstream we enhance our scheduling capabilityincluding the affinity in order to complement the scheduling capability of the upstream.And in terms of the network it is plugable.So that we can realize our multi control planes.We also have CNGini.That is an open source third party plugging softwareand it can realize our multi planes.So most of the time when we serve our clients we find that when migrate to the upstreamthey maintain their old mentality and they cannot accept the new ideas of the Kubernetes.They cannot quickly acquaint the technologies of the Kubernetes.So we need to ensure on one hand the product and technology,but on the other hand we also need to promote the Kubernetes.So I myself is organizing some online training and offline trainingto promote the cloud native technology.Hello everyone.I'm from Jindong.Actually in Jindong architecture it has begun in 2014and we began to use Docker and the OpenStack.So after its development after several yearsand you may notice that these years are the time for fast development of e-commerceand this give us posted new requirements for our bottom architecture.For example the scalability and the application.We also evaluate some similar orchestration solutions.So finally in 2016 finally we select Kubernetes as our solution.So since 2016 since that time maybe only very few peopleuse Kubernetes in 2016.So we are the pioneer to try the Kubernetesand during the whole process we also encountered a lot of problems.For instance we need to have customized thingsto follow the development,internal development processin addition the scheduling problem also the resource utilization.So later we have some optimization and some improvementuntil 2018in Jindong most of the productionthey are run on the Kubernetes cluster.So in present in our enterprisewe already stepped on to the second stage of Kubernetes.So now mainly we focus on how to improve the resource utilizationand currently because Kubernetes has its own schedulersbut we also have some research and development for customizationand we have our own algorithm added inin order to improve the bottom resource utilizationto improve the bottom architecturethings have been improved that we largely improve our resource utilizationand in the past one year in 618 festivalthe fundamental architecture of Jindonghas nothing to be added but actually our businessesis incremental and it is even much larger thanour previous business.Also one more thing that I want to share with you isrecently we have TrubalFS that is a new open sourcesoftware and it is a distributed systemand we use it to solve the problems encounteredin our production.So sometimes maybe non-state applicationis quite ideal but actually in productionwe need to store the data.So that means we havethe requirement for the long-term storageso based on these requirements and these demandswe come up with the TrubalFS project.Also weganked a lot from this project and since March it hasdiscussed this with Li Xiang and it is still at its initial stageand still much space for improvement but we want toborrow some forces or some power from the communityto strengthen this project and we invite more partnersand enterprises.我們對於CTO原生技術的這些手機技術關來說你們覺得這些不同的架構它的優點和缺點是什麼比如說私有員工有員還有混合員這樣還有多員平台because for the scenarios the private publicmulti-cloud and hybrid cloud because we haveour KPI clusters many KPI clustersactually not all the applications are on the cloudsome of them are on VMs so currentlywe have the container and the VMalso we have the host about the public cloudso from my own perspective if you have a large numberof businesses or if you have wide customers or clientsactually you do not have the rack in every cityso when you have such demand you need to purchasethe rack from the local suppliers and you needto schedule so that you will have the demandfor the public racks so that meansfor a long period of time you will utilise the private cloudpublic cloud and multi-cloud hybrid situationhybrid state about the selectionI think it largely depends on your businessthe size of your businessif your clients are in only one city or one provincemaybe only one rack is enoughbut if your applications are quite perfectand you can also put it on the physical hostso that depends on the size of your businesswell I myself think the public cloud is the future trendso in future maybe most of the enterprises would usethe public cloud or public cloud services more or lessfor private cloud maybe if you focus on securityor some and you will put it in the private cloudbecause it takes time and energy to put iton the public cloudso under such circumstances you will also considerthe hybrid cloud architectureand how to integrate the private and migratebetween the private cloud and public cloudalso for some meetable or elastic dataand how to realise the cost of savingto be cost efficient and reduce the costI think all these are the factors that need to beconsideration, taken into considerationabout the multi-cloudmaybe and present most enterprisesthey will select only one cloudfor example the Ali cloudbut in order to negotiate or have a backupthey may alsoapply or adopt some services in other cloudsfor example some enterprises like Twitterin North Americathey will adopt such strategyand they will not be bonded tologinthe approaches as I mentionedis according to your own business needsbut the overall trend is to develop tothe public cloudthis showcases the advantages of the cloud nativeand many architecture start from the layersof the cloud native and they can integratethe cloud provided servicesand capacityand it's kind of a universal APIso it can better migratethe state service and the non-stateservice on the architectureso the native cloud can release the pressureor the workload on thiswhile just Lisawith a detailed introductionI will add some pointsone for the gapactually Kubernetes has already becomethe standard specificationof cloud providersbut no matter whether youadopted the distribution versionor multi-cloud provided deploymentactually the cross-cloudcluster managementhas alreadyhas not been unifieddoes not have a unified standardso if the combination becomesincreasingly more complexthen the cluster managementcost will become higher accordinglybesides for the running timeon the cloudactually for the open sourceactually have already providedthe cross-cloud migrationincluding for the application layerit can help you to achievethe multi-cloudmultitask network connectionand for this applicationor collidingit does not need to worrythe communication networkon the lower layeractually for the connectionof various networkwe need to public ordispose all the informationon the public cloudthis is quite a concernand it's quite difficult to achieveand this is a relatively big gapbesides for your businessif they have the migrationacross the cloudthe data becomes a moreprominent problemthe flow is quite importantmost oftenwe choose to insiston the private cloudalthough some calculationon the public cloudor we insiston using onepublic cloudbecause we have the data qualityso we have a mass amount of datanot a freely flowon different cloudsso this is a gapfor the using of private cloudor public cloudor multi-cloudsso in the mixed architectureor frameworkcan you use in a trueprogram or solutionto address thatfor now the industryhave some also providedthe data flowonmulti-cloudsactually this isa big gapcurrentlywhen a previous guesthas made a very detailedintroduction actuallyI will skip some detailsand I would like to sharesome experienceI learned from a case studycurrently for theorder infrastructurecontains private cloudand the data updatingso most applicationsare deployedon the private cloudbasedoncubanetsand on this unifiedplatformsome basic softwarefor examplethe mid-warefire document systemand databaseimage databasek-value databaselike this lower layerapplicationsactually theynot achievedthey are not deployedon a cloudactually they deployedon our Disneyall I want to say hereis thatto put on a cloudactually it's notnot to be achievedon a single dayactually it's quitea long and hard journeyand we also need toadopt thataccording to our businessdevelopmentand other demandactually this statusrequirehigh requirementson your applicationsso just sharethe caseand for your reference那你認為服務meshes和沒有服務的沒有服務的這樣一個狀態是否對於企業的IT是成熟的呢為什麼或者是為什麼不這麼認為呢well forso smash andseriouslyI believethey havemany practiceand researchon this areawell I personally believethis technologiesis still in theearly stagefor the developmentand the maturityis need to be improvedwell actuallywe have done some testsI'd like to share thatfor our e-commerce platformthe performanceof so smash andseriouslyis need to befurther improvedand besidesfor theoperation complicityI think it's notsimple enoughto support the operationfor your referencewell I'd like to shareother experiencewell forseriously smash andseriouslyis actually will havea very deep study of thatfor so smashwe support moreincluding theis this communityupstream investmentactuallywe are theearliestinvestorin Istiocommunityactually somenew emergingquestionsand the scenarioare notthat sensitiveand I believethe evaluationis necessaryfor a large-scale companyand for theearlyattemptingactually the userwill quitewill quite cautiousabout thathe will firstmigratethedata to thatwell forlistinghave oneadvantage thatneeds, does not needto deployment oncubanetsso for somescenariowhen you dothe businessmodification onthe cubanetsyou have one moreoption thatyou can firstdo themicroservicemodificationso thatit canbetterconnectingthe serviceon the cubanetsand servicemodificationand this isadopted bymany peoplewell forservice matchesorservice listsone ofthe majorproblem in thatit is notit's notthat mature onthe open sourcein myas myI record thatlast yearit releasedthe6.0versionsomany cloudsdoesn't havethe unified versionof thator haveor does not supportfor thatfor last yeartheconnective was launchedand many platformsstartedto modifythe servicelist accordingto the cubanetsso for nowthe servicelistis timeto havemore developmentwell themetaphoservicemashes orservice liststhe ultimategois tomigrate theconcern ofthe originaldevelopmenttoinfrastructurelayerandall thesepressures orthe workthe effortsneed to bedown bythe infrastructureandultimategoof thatactually wewant toreduce thecapacity of themedaware totheinfrastructurelayerso thatwe canhave abettermanagementand havebettersupportfor multiserviceswell forone ofthe greatchallengesfor service meshas mentionedby collegeespecially forthe set cardwhen updatingthe set cardis not as easyas updatingthe othermedawarewell for thepublic cloudwhen you do theservice meshit's alreadymaterial onthe public cloudlike alibabaohawi hasalready done thatfor youfor thesecular concernis about theperformancefor theperformancefor thelong-term runthis willnot bethe alltryingdifferent waysto solve thatforwewant tofiltrecapacitydelaylower totheinfrastructurelayerand we usethepanelapproachto dealwithotherissuesif wejustcompare themtoasthe endperformancebut actuallymany public cloudalready usesador aslinesolutionsso firstif they canbringenough valuesso theperformance willdefinitely beimprovedfor thethird concernI believeis much moredifficulttosolve thatthereduce theflexibilityso if weuse asdkingpatterntheroundtimewithdifferentlanguageandapplicationsarequite goodandtheflexibilityisquiteunderstandingandwe usethesmashnessapproachtosolve thatitactuallysupportedmorescenariosandmorestructurelayerand wewant to do theflow controland if it canhave someself control on thatso thisfrom thisaspect for the longround it candominate thethe marketfor theseriously itactually it hasa broaderdefinitionas self-formationactually it'sthe definitionfrom theperspectiveonly oneaspectsthat is to sayif a service cancan meet a patternof pay on demandthen theresources can beabstructedso thispatterns can bequarteredso this patternincluding some othercompanies alreadyprovidedsome thisdatabasepatternsand thepayment pricingpatternswhich isthe traditionalpricingsystemsactuallypricing accordingto theCPU costactually costwhich isquiteflexibleand it canreallybring the costeffectivefor the usersand theconvenientfor usersthis isthementage ofsurnaceso forsurnaceI believewe needtoapprovepatternsand thatwill beeasierfor theusersto beexceptif wejustabstructedthesurnace isinto adifferentlayerwhichcannot beexceptedthenwe havea lotoflimitations on thator forthe broadersensewe canimaginethemodelsoto whichdegree canweabstructedtothatthesurnace iscommentexceptedthatand canbringmoreconvenienceforuserstheyhavemadegreatdiscruptionand I'dliketo sharemyusersforseries meshit focuses onthe infrastructurea dynamicand elasticand besidesit alsoemphasizesthe service should bemicroserviceand bestandardand serving the usersbut forcurrentlywe havenotachievedmicroserviceor serving the userswe stillon the early stageas Kevinactuallyseries meshconnectconnectthe physicalserversand the containersso we wantto usethe servicemeshes canconnect thesetarnalsfor examplewe cando the serviceidentifyingand userand workloadbalancingand wenotdo thesharecardwe justremovethemafterthe meshthen it canbe maturedand we willconsider laterfor the serverlessfor serverlessactuallydoesn'thave wide applicationin scenariosbecause firstwhen it isservelessand it requireshigherit has highrequirementshigherabstractionservicemeshesbut internallywe havethe serverlessservicesfor examplefor some newdevelopersbecause they canrun rapidlyand theyalreadysupporttheelasticityso I thinkthis would bethe futuretrendfor development非常感謝感谢大家的分享好大家感谢感谢