 Good afternoon, everyone. My name is Katano. I'm from NetApp. So for today's session, so myself Katano from NetApp and we have Mr. Masaki from Yahoo Japan. So this is a joint session from two companies. So for speakers, profiles, so Masaki-san can introduce yourself. Yes. I'm from Yahoo Japan. I'm a storage engineer. My name is Masaki. I'm my first year. I was assigned to applying storage operation team. There, I was handling the storage device operation and also I validated OS for the new components as well as the NetApp storage. Same time last year, I was assigned to private cloud team. So it's been since I involved myself in the open-staff operation. So now I belong to private cloud team. So I validated back-end storage and storage component and develop and open-staff sender and switch to related solutions. So my name is Katano. I'm from NetApp. So I'm responsible for telecom carrier and ID service industry customers. And I'm also responsible for pre-sales SE. And so I'm responsible for Yahoo Japan account. Again, I'm from NetApp. So this is the joint session and also just a community session. So the Yahoo is using the NetApp solution and open-stack. And they're using the open-stack in a large-scale commercial environment. So we would like to cover how to use this kind of the storage solution in a large-scale environment and also some of the points for future consideration for you. So this is today's agenda. So first of all so we're going to talk about how the Yahoo Japan decided to introduce the open-stack in their system and how they're currently using. And we'll also talk about the important points for the consideration when it comes to storage after that. So I will talk about the requirements for cinder in open-stack storage environment. So without further ado, Masaki Sunflow is yours. So from my side, I will talk about our approach for open-stack environment in Yahoo Japan. So as you know, in addition to internet portal-side operation, we provide a lot of other services. So it's important to offer the services swiftly and also while maintaining the stable operation. So I will talk about our insight. So let me talk about how we have decided to use the open-stack. Before open-stack deployment, we were using our the home-grown iOS environment as a private cloud, but there's a lot of things with that. First of all, API format is proprietary and it's quite difficult to develop the iOS environment and operate with the limited amount of resources and so therefore we have to allocate a lot of human resources for the operation which is slowing down the new functionality development. At the time of the release, of course, that devices are aging and there's a lot of advanced functions in corporate devices, but in order to use that, we have to create new environments so that we cannot really have better life cycles. Therefore, in the beginning of 2013, we started paying attention to an open-stack and a cloud-stack and we started the validation. Back in 2013 in Asia, cloud-stack was more popular. But outside of Asia, the open-stack was more popular. After the validation, because of the flexibility for the custom development and also the information availability, we have decided to choose an open-stack. Six months after we started the validation of the open-stack, we were able to release the open-stack in our open-stack environment. It's been two years since we started using the open-stack in our environment. Here's the reason why we chose the open-stack among all of the IR software. As I pointed out, there's a great momentum in open-stack adoption globally. Earlier stage, there's not much software in open-stack, but there's a strong user community in Japan. They're holding the seminar and workshop. Even we encountered some issues by asking the question to other users and inventors, we are able to solve the solution. Well, we come up with solutions, so we are the largest open-stack users here in Japan. And for some functionalities, and we don't have much choices for the hardware in the back end. In this case, we asked for extra work to the vendors, so the vendor prepared a unique driver for that. So, this is the verification to deployment. In Japan, we were able to release a development environment six months after our validation. I'll come back to this point later. By moving into open-stack environments, so through open-stack API, we are able to manage the data center. And in several dozen seconds, we can start several hundred instance. And after the release of development environment, six months later, we released a production environment. For the service side of members already used to using open-stack. So, several days after release, we are able to again deploy the service in an open-stack environment. The reason why we are able to deploy this environment in a short amount of time is because we don't have to prepare the unique API. In our traditional environment, we had to develop the basic functionalities and additional minor functionalities. And for account integration, we had to do everything by ourselves before. But after open-stack implementation, well, we can tap into basic functionality in the communities and for the some minor functionalities. We just outsource the minor functionality development to vendors. So, all I think we would do is that user-related functionalities for our private cloud. So, that's the only thing we take care of. So, we have completed a lot of the custom functionalities. So, now we are making the greater contribution to communities. So, if you look at this slide and using the open-stack platform, we are distributing this content. And so, this is the current services running on the open-stack. So, YafushaPan provides more than 100 services. And most of the services are running on the open-stack platform. So, this is not the new, and most of the services are already running on the open-stack for a year. Some of the incidents are constantly running. But an Olympic or soccer World Cup or general election, so, have a lot of access hike so that temporarily we enhance the instances. So, the reason why we can do is that because of the open-stack advantage which is enabling us to manage the large-scale resource falls. You can see the t-shirt. So, it said Bakusoku that explosively fast so that our slogan or vision is we have to meet at a rapidly changing environment. So, application release cycles for open-stack is more proactive for the environment changes so that we have a great affinity with open-stack philosophy with open-stack use slogan. So, here is the open-stack utilization. Currently, in Japan, we are running 4,000 compute nodes. You can see the sister user survey results and so that you can see the Yahoo's position here. In addition to computer nodes, computer core 96,000 for instance. So, you can see the 50,000. We have a result. We are among the top tier users so that we can say that we are the heavy users. The next is about our private cloud service service offering status. Under our environment, there are development environment and production environment instances. In total, we have about 50,000 instances. Maybe half and half of them are available. So, we have about 90% and production. So, those instances availability is 99.996%. So, this is satisfying the quality level. Currently, traffic density has grown six times comparing with the physical environment. The data volume capacity is 20 petabytes or about 20 clusters. This 20 cluster is quite liquid. So, there are some closed clusters or newly built clusters as well and overall they are cycling the life cycle or rolling the life cycle. So, there are 11 team members managing this private cloud. There are some members who have been in this private cloud team from the beginning but open stack is a combination of the network server stretch and so forth under the IAS technology so infra technology involved members also in a team development and operation responsibilities are allocated but other than these, we have appliance operating members as well. In case of the troubles on the open stack, the private cloud team responds however when it comes to the actual device problems, the relevant team members try to solve those issues. So, each team members are collaborating in closely. This is a user usage status. In Japan, we have about 2,000 engineers in all services that we are offering and all those engineers we offer the IAS environment in equivalent manner when the VM is a failure or performance degraded the office open stack operation team swiftly respond to those. Private cloud team as a part of the IAS environment proliferation activities we periodically hold seminars. This was cruised to the engineers in the beginning but recently we see the interest from other team members and they have started to participate. Maybe thanks to this seminar effect, this usage is preferring over the engineers, non-engineer employees as well. There are about 500 instances operated in a day. However, in a month's time half of this is said punishing or eliminated. Probably because it is a self-service environment, they deploy with some space of room and after the release they eliminate those. In comparison to the private cloud operation, the cost was significantly reduced under the regular specification you may not see a huge difference but the specification that we require which is about 2 core memory 12 gigabytes under Yahoo! So with this capacity the difference is huge. So rather than using the cloud, we have reduced the cost by 97%. So in regard to the operation status I would like to give you a tip. This is a photo of the Tokyo tower which is a radio tower very famous one in Tokyo. Under our open stack we use a stretch component which is consisted of the cinder and a small sized swift. So if we add all the disks add up to 377 meters Tokyo tower is 333 meters so our disk number is higher or our disk height is not higher than this Tokyo tower. So next I would like to talk about the benefit of using open stack. First of all the standardized API can be used by using the standardized API we are able to leverage the OSS around it. To date these areas had been prepared by ourselves so new IRS environment establishment or new development was thrown down. However recently we are able to allocate more people to the areas or layers closer to the applications. This is another benefit for the users as well. So leveraging the internet knowledge is very easy to use. The second benefit is the vendor not neutral resource utilization. Open stack has various vendor participation and they have open stack unique drivers as well. So we have more options on the hardware resources and that's able to have the abstraction of the hardware. Thirdly the data center abstraction. Open stack implementation to reduce the time required for the orders or delivery for the servers so like securing or whiling network design those things that need to be considered by the users are no longer necessary and that's deployment of the machines became easier. And not only limited to the virtual machine offering but orchestration deploy is possible so that is a big benefit of having the open stack. So this is a measure how we offer the service to the users. User use the virtual machine via the control panel that we prepare. Basically the resource management is done by the operation side and the user side released from the cumbersome activities such as to select the devices or the ordering this. So class start data center or hypervisor types of production or development or storage network they just need to choose out of the box that is prepared. So if they have particular requirements for the SSD for the application they send the request to us and we provide the service in order. So when the resource is reduced we are storing new clusters without any time-downs. So the integration interface that we are offering is as shown here. This UI offers the basic functionality of the open stack and the cluster based as usual resource statistic information virtual machine search functionalities are on here. Availability in zone based usage status can also be conformed. Initially you may not need the virtual machine but as you grow larger over time you need to consider the disaster recovery and distribute the data in both eastern west virtual machines. So this would be a useful tip for you in such event. So this UI allow the start-up of the cluster based deployment. So this is the hypervisor environment that we are currently using. This is a KVM environment. KVM environment is created using Chef. The open stack version based recipe is prepared so that the most optimized recipes is provided for the configuration. When it comes to the large size clusters, more than 200 hypervisor set up is required but with Chef only one command completes the deployment. Also under the KVM environment availability zone based flavors are prepared. So depending on the request from the user, a number of IOPS or reduction latency is sent to us. Therefore we prepare the devices depending on their request. Sometimes we offer the SST device or sometimes NVME offered. Other than these, in combination with the backend used stretch, the stretch's benefit should be maximized so the sender driver prepared by the vendor is also used. This is a VMware ESXI environment. Recently VMware ESXI used environment is reduced. VMware uses vCenter to allow the integration management of the ESXI under vSphere environment. If you use the VDS, VM access can be switched by the data center unit. The benefit of the VM where is the VDS is used is the VDS is used to reduce the dependency. VMware functionality has the VMware HA but when the parent server fails down on the but the healthy server can self-restore this so you can reduce the downtime of the instance. The storage is at the back and so instance image is stored in the storage side so there is no performance degradation. When you start the instance there is some cloning at the storage side for image so therefore we can eliminate unnecessary traffic hence we can reduce the I.O. This is the component selection consideration so in a company there are three key points we pay attention whenever we choose the storage component when we cannot stop I.O. Second, the time required for instance cloning thirdly for the vendor support. We cannot stop the I.O. That's very important for the business or service continuity. Some of our services include lifeline support so this is the significant point. Even there is some failure we have to continue the service in the front side so in the back end operation side we recover and using the failover and failback functionality and same thing is applicable to maintenance rolling updates so the updates while maintaining the services is an important criteria for component selection. The time required for the instance cloning is one of the most important criteria for storage selection. If we consider auto scaling again we cannot neglect this factor and vendor support is of the great importance when it comes to component selection. We have a numerous type of the services and there's a lot of the requirements for the service operation side. To meet the requirements by ourselves it's a lot of work therefore we need a vendor who can listen to outside requirements and a vendor who can collaborate with us to meet the end users requirements so that's why this is another important criteria. This is our standard implementation and there are three key points when is the shorter time for instance cloning in the back end storage and we are using the clone technology so we can start the huge volume of instance in the short time. The second is the storage installation per availability zone so we have a different kind of the power distribution in the data center so we separate the availability zone per power distribution line so in the same way we locate the storage so when the user starts the instance the user just pay attention to availability zone for the parent HP and they can set the redundancy between the instances therefore if the one power line goes down the other power line is live so therefore again that operation can be maintained. The third is the multi and back end configuration so we are pretty open to the storage in the back end so we using the appliance storage at the back end so for the stability and easy to use so the appliance is our best option but a software defined storage it can be one of our choices as well because of the technology advancement and we are the early adopter of the new technology so nowadays that SDS have some technology and so the other component and as one of the different kind of integration so that's quite important or interesting point so so far we have talked about open stack environment in our side so let's talk about why we have decided to use the net app storage outside of open stack environment we are the heavy user of the net app so we have lots of engineers who have a good knowledge of net app so that's one of the important reason so because some of the components are designed to protect the data such as Cinder we have drivers such as FlexClone which can leverage the net app functionalities these drivers we don't have to change the API in open stack so that can reduce our development workload so we have again the diverse services in IS environment so I believe the FlexClone it's a quite beneficial technology even before we move to the virtual environment we have been using the NFS common file storage the net app is quite strong in this area so Manila common file storage components is attractive choices from net app therefore and again we have to have very collaborative vendors for the continuous support so that we can have a room like the knowledge and experience so we call it core creation so just to symbolize a relationship with a net app so now I'd like to hand over to net app person to talk about consideration points for storage selection from net app perspective so this is a storage requirement specific to IS environment from a net app perspective so when it comes to environment storage requirement I think there are two perspectives perspective of infrastructure and perspective from IS environment or the infrastructure side of requirement and there are some general requirements high availabilities simple manageabilities and scalabilities and also well when the monster VM emerges we have to make sure that monster doesn't really have any impact as a VM so you have to have a final IO control in a large scale cloud environment and that data volume is quite high in storage and system disruption has a huge impact on the user side so therefore we have to minimize the system outage and as much as possible and for the users benefit so if we consider self service portals and what do you think is the requirement so net app has the in depth experience in this area especially we have a cluster architectures at the same time well self service portal requirements well we have to understand how the users are using the self service portal so the open stack users they create a tenant from the self service portal such as horizon so for number of instance and number of CPU core and storage capacity is set per tenant so here's what users do for tenant only done work is instance creation so to create in the backup using the snapshot and template upload traditionally these are handled by an IT department for the virtualization for instance for the backup sometimes the backup puts higher workload in the system so usually that backup is captured at night time for the job execution so the VM creation same if the several hundred VMs are created and started at the same time that puts the workload on a host and network side so if we cannot offload from the storage and that has a huge impact in the system side of performance so that has to be considered in the self service environment again we have to really understand additional workload which is potentially coming from the self service environment so as I pointed out so these are common work done by users and the VM replication and backup using snapshot and template upload by connecting this to storage and functionality so the VM replication so the VM can be replicated without consuming any resource and for snapshot without creating data therefore there is no performance degradation for the snapshot creation and for the data traffic via host and network so that it's possible to offload the workload in a storage side virtualization capability so all the data I can be offloaded to the network and overall infrastructure can be flexibly or stably utilized so this is about the KVM environment using offering the single driver for the KVM environment unified driver is offered this unified driver is leveraged to backup the snapshot offload or virtual machine cloning linking it to the flex chrome so the time constraints let me skip this slide next I'll try to talk about the VM way environment so this environment is when it is controlled by open stack when it is used by ESX ESX driver or vcenter it will be used so in case of the sender usage VMDK driver is leveraged to boot the instance or create the SSS vendor provided sender driver no longer be usable under this environment that can be a problem therefore the virtual machine replication would be done through the vSphere and VMOS snapshot would be created of course the traffic cannot be uploaded and the virtual environment plug-in tool that we offer virtual storage console the backup capability is offered by that so these tools are no longer usable under this environment so the self service portal requirement how we are able to do this basically VM way environment native functionality needs to be used therefore vSphere and stretch are linked which is VAI or NFC VAI or recently released Vivo those new technologies are leveraged that is a very important factor what is written on here is the integration details this is a network based linkage in case of the virtual machine croning or replication NFC or Vivo can be used for that linkage with the flux crone without degradation with the capacity VAI could also be used but VMDK used croning NF SK usage is better for this application and in case of the snapshot unfortunately VAI or NFC VAI function would not supplement this capability requirement so vSphere 6 released Vivo usage needs to be considered by using Vivo VM snapshot the VM crone but this is a new functionality so how you deploy this functionality needs to be done with a great portion lastly data creation offload so this functionality VAI or NFS VAI can be leveraged for this offloading so with this KVM vSphere environment have been explained but hypervisor depending on which hypervisor you use the functionality you need to utilize will be different so hypervisor the storage which has a relation or linkage capability needs to be chosen and you have to clarify the requirement for the self service environment lastly we would like to ask Mr. Masaki again to talk about his future initiatives so let me talk about the future schedule we would like to proceed with Manila the shared file system is demanded within a company as well and Manila allows the file sharing over the tenant and heat based on scale time you are able to create a shared file system so I think there are ways to utilize this and this shared file system has been widely used since the physical era and there are many it is easy to implement the service and there is no end to this utilization of this service and those users who were not able to migrate towards the I think this is a great feature lastly let me wrap up this session the benefit of the open stack is that the standardized API utilization and hardware resource utilization and data center abstraction and the stretch component provides the availability, efficiency and the safe environment so the vendor stretch should be used which has a proven record and know how and for the convenience improvement for the open stack environment the co-creation with the user and vendor is very crucial so that concludes our session if there are questions please contact us we will be here so please contact us directly thank you very much for your attention