 Hi everyone, we'll play a video first, yeah, hit it, sure SeaTrip, a travel industry leader and pioneer 1999 the SeaTrip dream team, James Liang, Jiqi, Shang Nampang and Fan Min founded SeaTrip Forging ahead on a 10-year journey bearing great dreams and hopes Any talented people joined them and the team quickly grew in its first two years SeaTrip quickly fulfilled its great early promise Incorporating biz express and acquiring modern express business travel services and Beijing high-end air ticketing company a wide range of teams contributed laying solid foundation for the company's continued growth in 2002 SeaTrip exceeded R&B 100 million in monthly transaction value In 2003, SeaTrip made its IPO on Nasdaq and set a three-year record for a one-day share price gain for an IPO After this successful listing, SeaTrip took a leading role in the market and began to revolutionize the industry In 2004, SeaTrip first introduced its vacation packages supermarket Establishing a new travel trend in the market Customer-centered service philosophy has been the key factor that helped SeaTrip excel In 2007, SeaTrip moved its headquarters to 99 Fujian Road, Shanghai 2008 Prime Minister Wen Jiabao visited SeaTrip headquarters in Shanghai Expressing hopes the company would continue creating a brighter future In 2010, SeaTrip established a major new call center in Nantong Jiangsu province To enhance its presence in the leisure tourism market Continuing to set trends in China's online travel service market and laying the foundation for future growth Travel with SeaTrip in the decade to come A little bit of highlight promotion for SeaTrip When we play the video, we have a meaning for it, not just promoting And the team will talk about that during the presentation, why we play the video Relating to the cloud technology, today everything about clouds Let me hit my thing, this is a agenda, but we will go through it First one, come back Problems we are facing, I just want to talk about that one As head of technology for SeaTrip in the last 10 years has a tremendous growth In the next 7 years I mentioned in the morning and business growth 10 times So right now we are adding new business lines In the last 6 months we are adding 5, 6 new business lines, side doubles every year And business wants something yesterday as a technologist you guys are doing And that you heard in the past many times Then side availability must go up Infrastructure costs must go down So those are the KPIs, I have to keep high, it's the high level When I say side availability goes up is that 10 times more reliable than today Infrastructure goes down per order Today we have a dollar a month per order cost And I want to drive that down aggressively Cloud going to help at 2000 degree, we will share some showcase with you guys Yeah, call center I mentioned Next one, the V desktop I touched in the morning So the reason is that why we do this, one is cost reason And I mentioned that and there is a bunch of other reasons I didn't go in detail in the morning Here you see fast deployment of the physical box deployment versus virtual box deployment Which will be much faster with the V desktop Massive configuration changes What I mean is that if you know managing the data center and you will see Some OS upgrades, browser upgrades from IE to Chrome Now if you're dealing with 100 desktops it's not issue If you're dealing with Southern become headache Now if you're dealing with 10,000, how would you be able to do it? Take a year to upgrade OS because business you continue, right? You guys stop answering and let me finish the OS upgrade, you cannot do that So that's the with the VM on the back I just swap the VM like one second You got everything new, right? So that's physical versus virtual, that's the flexibility we're talking about All right, so the another thing challenges that I mentioned Three call centers physically distributed to different regions Now we're creating more call centers Something we already working right now is called SOHO Remote call centers, some people can call answering call from home That requires technology to do that And now we have many agents working from home And if we have technology ready for them, that's increased call age workforce Also lower the cost because they don't have to physically come to the call center answering the phone Some calls can take from at home is okay So those are the reasons Then the private clouds today we're going to get more details in why we're doing that One thing I mentioned 10 times growth in the quite large hardware Computing power, computing capacity And more machines when you manage that too And today if you look at the data center, it's a typical, okay I look at a couple data centers across industry, e-commerce side Server hardware utilizations at call center, any given call center Without cloud technology is very low And that's the number I can call it and you can use it for your reference And some majority of call centers, server utilization is only about 25% So in other words 70% the power, the capacity was not being used And because the technology used to build the call center and how you manage it So there's quite a depth details on that But I don't point is that is if you don't have a technology to manage the call center Cost going to the cost increase like this And so there's a call elastic computing If we can shrink and expand as needed as the traffic grows as business grows That would be best utilize the resources in the IDC environment So the businesses that you are talking about If you talk about new businesses they don't have much traffic But it's a new business you have to work on Also you have to do active-active because achieve high availability So two servers minimum however the business Today's server is like a four core ACPUs and it's kind of a commodity already And they don't, new business doesn't consume lots of computing powers But you need to give a server to them So put new business stack together as into one server That would be reduce the hardware footprint dramatically So that's the thing we have to use in some virtualization cloud technology First deployment I don't, you guys know already So it's provisioning VM stuff Right now we're adding servers every week like hundreds servers Then how you can quickly, I don't want to add servers like take weeks to add I want to take a minute to add servers into the data center You have to have some technology to do that Otherwise manually do installation it's not possible at that speed Alright, so those are some reasons we like using cloud technology Especially is open stack is we using in many areas, both areas And now I introduce my team, they working on this technology for a couple years now And have, they want to share with you technical know how they did it Let me introduce Simon and Yiting at the same time So both of them, in my team is system engineering And they have creating a group working on the cloud technology in those two areas They're getting into technical details, okay, have a phone Okay, thanks Thank you very much Okay, thanks Eric to highlight the problem we are facing And also the approach we are going to take My name is Simon Chen, I just mentioned And I'm the director for system engineers Our department is responsible for carry out or implement the strategy For Ctrip's cloud computing So once the direction is set, the strategy is set The next thing is some technical decisions So we have to, from my department we have to make some tough decisions First of all, why open stack? Why open stack? We take, we took quite a long time to evaluate a couple of options A cloud stack is one, and also whereas others And when you went to make us, made our mind, is the IBM joined the open stack camp That's the beginning of the year, so make this decision a little bit easier Before that we do the both tracking and try to see which one is easier And also if the data set is more or richer So after IBM joined the open stack camp We designed the open stack technology As a very big active community And meanwhile it has a relatively richer module set And of course it has the heavy endorsers like IBM And also we noticed it has some mission critical deployment already Like people have already done that So we think it's sort of a proven technology So we adapted the open stack as our technical, the cloud computing platform And then we have to decide, okay first of all we have to go to the cloud cloud Why we go to a private cloud Actually this is the relatively easy decision In China as you guys already know, there's no alternative The Amazon is sort of too far away They talk about coming to China and keep talking, keep talking And we don't see them arrive And Microsoft set up something, but to us it's less proven So we think oh bad, bad, we just go with the private one That's why we decided to go to the private cloud And the third one is why we build a in-house cloud management platform Instead of buy one or use some ready technology or product The reason is we try a couple of combinations But we all run into the problem with the integration With our existing tools and platforms So we decided we'd rather just go from the ground up Build the one ourselves It will manage both bare metal and VM provisioning And also other management tasks The third one, strategy has been set And the technology decision has been made Now it's execution time So all of the execution depends on the team We have to make a decision If we want to take time to build a full staff team Or we rely heavily on the community And we have a small team So we designed a small team Because one thing is the open stack engineers are very hard to come by We are still hiring, but it's not easy, you guys know So we decided to have a small team but rely on the community We have right now 15 engineers, very talented Very dedicated, highly dedicated And as you can see we have within a year Delivered three product lines The bare metal provisioning management The VM provisioning management And also VDI management platform So our team is very dedicated If you want to put a face to the team Here's the guy And Yi Ting Wu is the senior manager for our cloud platform He will go through the details Tell you guys what we have done And what lesson we have learned So here's Yi Ting Another tip, here's the mouse Later on if you have a wireless keyboard and mouse You can use it as a click, very handy You can try at home Thank you Simon and Eric Eric and Simon introduced a lot of business requirements And also our cloud strategies Now I'm going to give you more detail from technical perspective So I think Eric during the keynote We introduced a lot of numbers about our core centers And right now we have more than 10,000 core center agents Besides those later How was it? Sorry Hello? Hello? Besides those later I'm going to tell you a story It was beginning in the early this year And one day Eric and Simon will meet together And Eric asks, hey Yi Ting Is it possible to build a VDI solution for our core center? And I say, yeah, we can try And so we get started and do a prototype And at that time we also heavily developed Based on OpenStack And finally we find that OpenStack is actually a very good back end for VDI And we start hiring and then we build a small team And today we just load all the 1,100 things In our core center and both for some interns And the goal is to deploy 1,000 with desktop instance in our core center And also we start a new satellite office Which will have 300 agents So we will also deploy the desktop layer And there will be several hundred miles away But we will have dedicated network to connect our headquarters and the satellite office So I'm going to give you a demo To help you understand how we set up this demo And also thank you to the staff They have a lot I have a small sync line here And then you will connect to a switch And that switch connect to our laptop On the laptop here The sync pad and actually he run it open to end With DevStack and also KBM And that Windows 7 is actually We just started playing the slides and also the videos So that sounds fantastic, right? We already being in our VDesktop Now I'm going to give you some demos here This is our login window It's pretty simple Our core center agent just enter their username and password That password actually is our core authentication Which is use Windows ADS backend And then you just login Let me type in the password Here it goes, it's connecting Now you see the video, right? I'm going to play the video again Now just for a second Okay, I'm not going to play it again Now We also have This is the dashboard based on horizon And we added dashboard layer for the VDI And actually you can see a VM pool Which is over here I will give back bigger Yeah And you can see some cool information here The MAC address about the sync line And also the IP address And we use the demo user to login this VDesktop And we also have some very cool dashboard here This, those data are not accurate Because we don't actually have so many VMs running But to give you an idea We will use this dashboard in our production And also I would like to Level something really interesting Now you see that we use the VMC And you see yourself, right? It's recursive, okay? This is really cool And I think when our customers Or any call center agents They have problems And our help desk We will just login and see the scenarios What else happens, right? As soon as possible Now I'm going to go back to the slides So let's give you an overview How we put the desktop on top of OpenStack And we simply add sync line And for Keystone we use the Windows ADS backend And actually there are a lot of development based on Keystone So that it can support some group policy And domain, OU, something like that To support the VDI So for the sync client It will first try to authenticate with Keystone And then when it is passed authentication and authorization It will go back and say hello to the Nova API Give me a VM I need this template and this spec And the Nova API will try to Talk to the scheduler And the scheduler will find the best match VM, either it will be in our existing VM pools Or it will schedule a new VM Across our computer nodes And also we have NAS storage I will give more detail later And this is to give you an overview about our stack And in the computer node We are actually running In local storage with two 200GB SSD And also 160GB memory Two sockets, 8 cores CPU And for networking we have that And also internet network And we use NAS And it actually mix with Sata disk Which can have good performance And also cost balance For the ARM sync client This is the very important thing We have different spec for different scenarios Or different requirements It can have one CPU or two CPU Or one gig memory or less As for the computer nodes We are actually running on Ubuntu And yesterday I actually See the CEO of Ubuntu We have a small chat It's based on KBM And normal on top of it And in the network side We are using Neutron We open VStorage and VMode And currently it's quite enough for us For the VMode So for Swift We will use it to backup For the instance backup Or snapshot Or templates in glance For the Windows roaming profile And some folder redirection We will put it into our storage In the backend And for the ARM sync client We are actually running also on Ubuntu And we use stored for the large scale Deployment and management And obviously it's using the spies So you know our core center Is actually running 24 by 7 And on shifting All right During the shifting period There will be Thousands of users Trying to boot their VM And log in and log out So one challenge we have Is how to deal with boot storm We try many solutions like Fiber channel over Fiber channel And it turns out that It can have good performance Really expensive for a large scale deployment And easy to become a bottleneck And we also try local storage It simply doesn't have Enough IOPS for us Finally we will find the best match Which is the commodity SSD And our performance Testing data show that You can link Chrome 1 VM In one second And we put 50 VM At the same time It takes 30 seconds Really fast And good enough for us And we also have a program To simulate the workload Of our core center agents And the testing results show that We can actually log in 60 Windows 7 VM In single computer node So what about logging in and log out storm We try to put our roaming profile And folder To redirect into the NAS And the reason why We don't put the user data And profile in local SSD Is that it's quite expensive Some users will have several Personal data And also it will become A single point of failure Once the computer is down There's no way Or you need to take a long time To recover the user's data So for user profile Actually it's quite small For our scenario Like 10 MB But for user data It will be several hundred MB Or several gigabytes And for the first login It generally takes 20 to 30 seconds That means Windows also Need to prepare for desktop And for the second run It will be less than 10 seconds So what about live migration I think live migration is really important For maintenance and also for Prior really good User experience to our customers Thanks to KBM And actually with golden image Both on the same On different computer nodes With link clone You can just migrate incremental To another computer node And our testing data shows that With one gig memory Two gig memory And one gig incremental It actually takes 30 seconds And since we have the same client In the user side In the user side We can make our live migration Transparent to our customer Because when we do live migration And the connection between the user And our computer node Will be disconnect And we have reconnect A mechanism in the same client In the same client we will try to Talk with Nova API Hey, what's happening Where's my VM Okay, you migrate to another host And I will connect to that host For access And that generally will take Like 3 to 5 seconds And user will just see the brand And see our logo reconnecting And it will recover So this is also really cool And good enough for our scenario So what about user experience You know we already deployed 100 same client In our core center And some user sometimes They will complain Okay, the VM is not as fast As my desktop So how will you measure Your performance As our core center is Actually our agents get a call And then they will Talk with our customer And finally we will put the order Or change the order in our offline system So the page speed Of offline Is really important To measure our user experience So we have a real time monitoring And you will see How many times we spend On the dome landing And how many for DNS do cut So I'm going to end Here for the V desktop Because we also have Another cool demo to show you Okay, it's about bare metal provisioning So generally when Side ops put the bare metal On top of the rack It will take several weeks Before it really takes Transaction in production You will see that Side ops will power On the bare metal And you will be able to Provide provisioning with them And then waiting for networking Engineer to allocate IP address And the assets Will need to do the baking And also hardware configuration Such as ILO configuration Rate Upgrade the firmware And also configure the BIOS After that you can have OS installed But especially for Windows Even you have OS installed You still need to do a lot Of post install steps Such as patching Activating licensing And Configuring IP address Or even a nick bond And finally you can switch it To the production video And you can start all your applications On that bare metal So at the time When we tried to build The provisioning system We take We look into the bare metal provisioning There is a driver existing in the code But it does It has very limited feature It only takes the PXE And there is no way to do Post install Maybe you can do it With puppet or a shot And at that time We have already been working on laser Which is a Public labs sub project And with laser you can do a lot more Also we also customize laser I will give you more detail later And with our Normal laser driver We actually can take over this auto discovery Baking Hardware configuration OS installation And also the post install And right now we are also working with our Networking engineering team to Do the video switching API And actually they already have the API ready And we are working on testing Now I'm going to Switch the demo to my Mac Okay I think you can read it on the screen Okay So this is our customized portal And This is actually a very cute logo So auto discovery Now you can see several parameters In this table And this one I think we can click and see This is From Huawei You see a lot of details here A lot of details Memory, nick, mac address CPU And also we have several HP parameters here Like 360 and you can do search 360 And here you go Now I'm going to provision one parameter here Let's say we want to play with this one 360G7 Okay And then click the provision button Here is the UI and then you can select Baking How many hours you want And what's the rate you want to do Okay, rate one And what's the OS? Currently we support ESSI 5.1 And also Ubuntu SandOS Including Windows 2K8 Let's try Ubuntu Oh sorry, we forget to select The networking configuration So for the ILO we have several Network here And we choose the BMC Which is our provisioning Our ILO Oh sorry, it's our ILO video And then we also have a Production video here Let's go down And then just click Now you can see the Job is doing Here, right? It's launched And you also see several You cannot see the IP address Because It haven't go to the Sorry, go to the neutron And allocate for IP address So that's for the ILO IP Now you can see many jobs Already being done here We can click and see What we have And here is Where you submit And what's the IP address Being allocate to BMC or the NIC For your production video And also here is the detail log Baking firmware Sorry The screen just log And also for For the BIOS And pre-seed This is actually in the IPXC And finally you will see Something that APD gets And also Configure the networking I will also show you a Windows Let me find the Windows Here Sorry, I think the Screen doesn't work It turned on and my Mac Actually auto battery Sorry But I think we We almost have them all down Okay This is my Think client So it doesn't I think it's really cool And it's really fantastic And I really want to take this Opportunity to say Thank you to my team They might not be able To see the live videos But I also want to Say big thank you to them They are very tenant And also Working really hard To make this happen So again, thank you I think we We haven't ended yet Sorry We haven't ended yet And I see the slides Thank you Is it possible to Okay We also have one more So it's a lot of stuff We've been doing I'm going to bypass this one And those are A lot of details here I'm not I don't have time to cover And yesterday actually I'm talking with The public lab team And they Actually working on I think it's almost the same thing They want to bring Nova inside OpenStack Sorry Bring laser inside OpenStack And I think we can We maybe can Collaborate together And bring Maybe sometimes we can Become open source And that's the one thing We really want to do So some challenge here And now for VM provisioning We have lots of VMs And the QV farm, Dell farm have more than 1000 VMs And also 1500 VMs in production And I think OpenStack Have a lot to help us To Speed up the This cycle Because The developers may need Dell VM And they also need Some feature testing Lot of performance testing And also the final staging Before you really go to production We have all those environment Based on OpenStack And it really helps a lot And also for the The builder Actually it's a Distributed build system It's on top of OpenStack And actually we are Doing that based on Jenkins And Jenkins has Actually has a cloud plugin Which could do auto rescale So If we have too many build requests And we will just scale out And when the queue Becomes smaller And we probably just Decum some VMs And return it back to our pool So for VMware support You know we are already Using VMware in our production So We need to have a feature To help our Operation to import Those existing VMs And ESSI As for the VMware driver We add some enhancements And also we also talk to IBM developers And maybe there will be some Collaborate For networking We use in Neutron And also we support DB switch And distributed virtual switch And also we add Customization specs support The template measurement Will be Help operation to Import the existing template In the data store So for scheduler filters We have a very unique Requirements Which is that we have some ESS hypervisors Which they use in local storage So That doesn't make much sense To create a cluster And it will block you Because once you need to move One hypervisor in or out It takes a long time So we just create A hypervisor folder And put those ESSI Under this folder So we add a schedule filter There And also for data store For different Tenants We can have different data Store Regular especially So that can be used to Find the right one Data store for you And we also integrate With DNS service Windows AD as a service Including the Zavix API I want to take some seconds Here to mention about our DNS We actually For the internal DNS We are using Windows DNS For a historical reason And our Internet DNS will be combined And we also know that There is a project called Desanet So we will We will work Working on And bring our DNS API And add that driver Into Desanet So One more thing is That we are Working on Bringing our Or migrate our Excessive VMware infrastructure Into KBM Some reason Like cost saving Lots of license saving Actually And also we can As a website You can do HGA inside application layer Instead of in the OS layer Or in the infrastructure layer And we also Want to Have more open source And a pan This month We just start Building our New staging environment 100 VMs Could be more On KBM And we are Just starting Migrating our QV farm There farm And also our Open-style We will add features So that we can Support both KBM and VMware together And we will see More migration In production Next year So I think That's all Thank you Thank you for joining This session And thank you for Your support