 Okay. Hello everyone. My name is EJ. I'd like to represent industrial technology research institute or so known as E-Tree. E-Tree is a non-profit organization that working as a bridge between Taiwan government, industrial as well as the economy. The way that we work is that we take government funding about 50% of them and then we work on interesting project and then transfer into prototype or technology that that can transfer into the vendors or industrial companies. So right here that we have our own distribution that we are going to work with the Taiwan government so we can push the open-stack inside in the Taiwan area as well as the Asia area. So I would like to show a short video so to show you regarding just a simple introduction for our distribution. Get this thing out. Nowadays lots of internet services are provided via cloud but cloud systems can be costly complex to set up and difficult to manage often providing an incomplete solution that requires heavy hardware software resources and manpower. Now there is a better way to build hassle-free open-stack cloud systems. E-Tree open-stack distribution this is IOD a fully integrated easy to manage one-stop solution that saves time and cost augmenting open-stack with powerful functionality. IOD is a dynamic open-stack enabler for carrier grade deployments it is made up of disco Peregrine and PDCM. Disco is a high-performance distributed block storage solution for Cinder that offers higher performance than CEPH integrating different storage and backup solution. Intelligent data deduplication reduces storage requirements by eliminating redundant data. Additionally disco offers HA, self-healing, N-Way replication, fast volume clone and data center to data center backup. Peregrine is our optimized networking plugin for Neutron. Fast failover pre-calculate backup path and immediately deploys it when a link is down. Traffic engineering dynamically calculates the packet transmission path and balances the traffic load on each physical link. Diagnostic UI provides physical virtual topology and traffic load, VM traffic load and traffic analysis. Commodity Ethernet switch uses OVS and Ethernet switch to provide SDN features making it cost-efficient. Peregrine is L2 fabric architecture and able to achieve optimal load balance of all the physical networks by dynamically calculating the packet transmission path. Peregrine is able to redeploy packet transmission path when any of the links or devices have failed. By applying centralization control architecture in fast failover. Peregrine as a hybrid SDN solution supports commodity Ethernet switch and virtual open flow switch. PDCM stands for physical data center management. It is our hardware device monitoring and service management system. Fire a user-friendly interface it provides continuous diagnostic monitoring of the entire system. Performing requested commands easily and providing built-in event management. How to save even more time and money? Bampi is a time provisioning software to save time and cost. Offering deployment automation, time and labor savings and ease of use. Without Bampi auto deployment from bare metal takes one person up to a full 288 hours. Incredibly the same task with Bampi takes only 1.5 hours per person. All IOD ensures high availability. IOD is a better open stack distribution offering a true one-stop cost-effective powerful and easy to use solution. I guess we don't have much time to raise the mouse. Sorry about the technical difficulty we are trying to get the video out for this video. So I'm going to because we have a next speaker that also have 20 minutes so I only have like seven minutes left. So I have to skim through some of the things. Maybe make it really quick. So what is each open stack? Here you can see that most of the core of the open stack are still there. But what we doing is that we offer this disco block device through the center blocking and a pair of green controller that add in our SDN. So we are defining it working through the neutron blocking and also the PDCA for monitoring. And overall we also do the HA for every component inside open stack as well as for our disco and peregrine. So I don't think I have much point to we don't probably don't have time to cover all those what is missing in open stack. We know we all know that the open stack is great. It's developed very fast but why it is still meaning something. In the early day the deployment is a hassle. Just two deployment all in one take you a lot of time and it's a stunning job. But nowadays the deployment become easier because a lot of company they have their own installation GUI. So that problem will go away. I think it's go away right now. But the problem is that for operations to your pain because the instance doesn't run itself is to your need people to run your instance. So that's why OpenStack Foundation they have this certified OpenStack administrator. So our job here is that we not only provide the system itself we also provide operation utility required to operate the cloud. So I'm going just to jump through a little bit and show you the architecture the OpenStack stack that we have here. So most of the components are the same. So except that in this portion we have our storage through the cinder and the low portion that we have our SDN solution peregrine that control the ethernet switch. This controller is able to use SNMP to control the ethernet switch or as well as the OVS switch. So I will talk about it if I have time. And then we have this PDCM for the monitoring. So I just have to quickly go through the some of the components. This is the disco the distributed integrated storage and comprehensive with configuration data protection. And this is architecture. We have the Compuno that has the DMX client that talk to that provide the VM the violence to the VM and you talk to the NEM node is the metadata repository and also all the payload request is extended to the data nodes. So what is the characteristics that we provide for our disco solution? First we have thin provisioning we have HA support and it has transparent data protection and fast volume cloning. I would like to highlight the fast volume cloning. In OpenStack usually when you start a VM instance you have to copy it over to a local storage or start it right there from the bottom. But if you want to start a VM with like 100 gig of the system valid you may have a hard time just to copy that things over your Compuno or to start a new instance. So we have this fast volume cloning that you can copy and write a new volume in a couple seconds so you can start your own new instance really fast. I will going to just come through this is just a feature in disco that provide the remote backup solution. This is also very important because in a cloud we have a lot of images that actually duplicate a lot of windows and probably Linux image that are all duplicate and also for example if you run an email system your attachment may just send to a lot of everybody and all those attachments are duplicate. So our system can take all those duplicate blocks and then just point it to the unique one so that can save you space for more important data. So I don't think I have time to go through the diagnostic UI but I will just show you a point that's that this UI is quite sophisticated that you can when you have experienced any problem you can see that there is the IOPS screen that show you the current performance. You experience any problem you can just go through UI look through the problem for example which VN use which violin and which data node and which non and in the non you may look at the rate status and maybe there are some discolors we got replaced and now it building so you the performance suffering so all those you can get it from our UI. So far is our off-lash solution that is actually built on top of our commodity hardware and as you can see that in the front we have this 24 SSDs so in total we can do one million ratings for 4k IOPS through the network so it is quite the performance is quite good and we are actually provide a lot of very good features in this product and if you want to have a good performance in some of the database you can consider use this product and we also trying to integrate the disco with so far is a work in progress. The global wearing wear leveling is a very important feature because over time the flash disk got wear in different level so it could be very different like in this case over time you become some of these wear a lot and some of these wear a lot little so eventually one of these got corrupted or use up then the whole system render isn't use that so this this one is very important feature so you can get our your system running longer next I would like to introduce you about this SDN solution from this we call it power green so we are a major member for the open data that's a open source and we contribute our code called SNMP for SDN the good thing about this one is that you can use the SNMP to config the traditional commodity Ethernet switch so you can just use it just like another OVS switch so what you what you can do is that you can set up flow through the SNMP and this so your controller doesn't care about is a traditional OVS so you can save money and also you your performance is quite good it's just up to par with the regular regular switch so there are some great feature inside the pair of green for example you can just use commodity Ethernet switch which is much cheaper than OVS switch and we have fast fail over so we pre calculate the fail over path so when there is a link down a switch down you will fail over to the pre-calculated path within one second and you will calculate the next one so your network is is got protected and traffic engineering that can low balancing the traffic low on each physical link and we have diagnosed with UI which is quite good I may show you some snapshot later so this is talking about the fail over so if there is a link down you'll fail over to new configuration and for traditional switch it just send SNMP SNMP through the component in the open data and then reconfigure switch so the manager UI here is as you can see that you can see the topology as usual and then for each link you can see what is the V then run inside this link and for each V then you can see what is the VM pair it opens stack that running through this vNet usually a vNet is is once network inside a tenant okay for each and then for each V and link you can see what's the what's the communication inside for example is it running UDP and what port or TCP so you can use it to cut down is the what is the problem for some kind of network problem or congestion PDCM is our monitoring tool so it actually have great features and is missing currently inside the default open stack so what it can do you can do a lot of things I because that's time I won't go over them here the good is look something like this you can look at a lot of things for example the CPU the fans and the temperatures so I'm going to move to the next subtopic regarding this rescue architecture so in today's architecture that I probably used here about the Intel's rescue architecture they are using a different energy there they use the 10g or 40g Ethernet to connect their common servers in the rack but here we are using PC PCI Express instead of the Ethernet so the good thing about Ethernet is that it's cheaper cost much less and also it's quite fast that I increasingly usually right now the mainstream of Ethernet is that 10g probably go up to 40g but it's very easy to do 64g inside the PCI so I think it's a quite good alternative to for doing the the fast and chived networking in the in a rack so from here that we have a lot of the traditional host that we have each host have its own CPU memory and disk but in the PCI based rescue architecture we disaggregate the rack into a CPU pool, DRAM pool, disk pool and nick pools and then we can create dynamically the the so-called software defined server that server can compose compose in different configurations for example you can make this server is suitable for the CPU intensive or memory intensive or network intensive workflow and the other thing is that because of this architecture we can make the server easily config and also very high density so this is one of the prototype that we do basically here is that we have this NDP to to mapping the virtual function of a physical function in this case we have Ethernet that has two virtual functions and we have them the managed host to mapping it to one of the vf1 to the left-hand side to one of the hosts and the other virtual function to to the other host so both of these hosts get one Ethernet device actually is that reside in his box it's actually reside under there the SIOV device that provide the actual device but we mapping it dynamically into other hosts so that's the idea and we are going to work with a Taiwan industry of vendors hardware vendors to push this architecture for the next generation of data centers so I guess only have three minutes left so I would like to conclude this portion of the talk is anybody have any question you know that I will invite a next speaker Dr. Xiao thank you good day everyone welcome to the to this session my name is Morris Xiao I'm the duty director general of data institute which is a part of institute for information industry in Taiwan Republic of China in this section I'm going to share with you our work in building a big data analytic platform and its application and the platform itself use several of the open-stake technology to provide cloud-based service which I will describe more detail later the the funding for this work is supported by a ministry of economic affair in Taiwan RLC okay after war is moving from energy economy to data economy and it's also gradually moving to the knowledge economy and there's one thing in common between energy economy and data economy in both economy that the value add coming from not just getting the raw material it needs to do a lot of you can call it a processing refining or whatever and applied to get the value out of the raw material and the more you refine it the more the higher value it get and there's one big difference in energy economy the resource is limited oil it get depleted sooner or later however data you get more value out of it it will more directly generate and it will explore you will never exhaust the the data and the explosion of data generate is commonly turned as a big data challenge which actually is first coined by a gardener research and is commonly used by the industry or academy today what it exactly is a big data they are full property company use to describe big data due to the time concern I probably won't go through this chart I believe many of you probably know this already the big data there's a challenge challenge normally means for engineer or technology like me I say I would like to solve it for business executive they say oh this is a big business opportunity there's market opportunity I want to get so in our organization we say okay here's a challenge there's market opportunity what we like to do and that's why we are building or we have built a big data analytics platform which we call a bit saw and the bit saw has two layer so at the bottom layer is a processing engine essentially we get many of the open source component including how to spark and and other things and integrate together and put the process engine and we also add a few of the utility to make it easier to install easier to monitor or manage and on top of that see a system is useful only if you have data and in big data area you need to go through getting that from the data source and then doing data orchestration data curation data processing data management and then analyze it to get insight and when you have insight you can apply in different industrial sector okay so this is from the open source mostly this is homegrown we are building the technology or to extract the data to enrich the information to be relation among the data set and create metadata and also to provide security governance and then have a good interface for user API for for data scientists to be able to browse search or use the data and so this is the architecture and so this can be used by any company or school to set up and run it and one problem in Taiwan is many of organizing are small and medium company or organize doesn't have a sort of large IT step organization for them to run this or the maintenance is a big problem and that's why we also came up with a different version or multiple expanded version which we use many of the open-stack technology to provide the big data analytic service which means it's a cloud-based big data analytics service so this is architecture so this part and this part are the one in the previous chart and in the processing engine we develop or implement five new component including identity management class management repository object store and management and those components interact with the technology from the open-stack and through the RISC for API and then the original processing engine will use this as the interface to interact or use whatever function provided by the open-stack at the similarly we have a bit so serve a certain client which essentially is to do the provisioning class management and configuration function those are also interact through this to get the function or to use the open-stack okay and this is example just to show how the business edition work in in in application and this example show we are you implement a recommendation system so in the configuration stage you need to create an account to allocate a resource and also to decide what where to store the result and in for the data scientists or data engineer during the analytic application stage then it needs to do to specify the data repository and then to develop the logic to get the data process data and analyze the data and also specified output and from this is done the output of the analyzer result will be stored in intermediate stage and it can be used later on by the business person or data scientist and this is just to describe what of the open-stack component is used in each of the stage including keystone, heat, NOVA and SWIT okay so the analyzer result will be stored in a recommendation knowledge base which will be used repeatedly depends on the application scenario in this case is recommendation so when you want to recommend certain product or certain thing to customer or to whoever then it will search or retrieve from this knowledge base and use it okay so in the remaining part of this section I will give you some use case or success story or application or company using the BISSO system in Taiwan one is in smart commerce the other one is in smart manufacture okay the smart commerce is the one we work with the body shop a body shop is the United Kingdom company which has more than 2,000 store in over 5 or 50 country 50 country in Taiwan the body shop has more than 400,000 members and it has a very comprehensive member system keeping track of the the member information and the body shop wants to essentially find a way to improve the membership service and also to increase their revenue and profit and so far the one is to have a personalized recommendation system and through the cooperation working with them we decide on two solution for them one is a personalized recommendation the other one is also associative analytic and this one I think in many e-commerce side Amazon or Netflix already also have done that essentially is try to base on the person profiling or property to recommend other thing and this is its kind of combination of if a customer buy certain things what else maybe he or she will buy at the same time okay and to develop these two solution or to solve this problem you need to go through several stage like this and technically in each of stage we apply several or develop several technology including identity matching, data cleansing, feature extraction and in the analytic stage including the customer clustering, collaborative filtering and content based recommendation and for this we can develop the frequent patent mining and high utility patent mining and after this stage the result will be generated and it can be used by the in body shop case they are used in the the physical store during checkout and what that how it works is up you check out all then the screen will show up said what item get bought by the current customer and immediately system can try to match and do a recommendation on the screen and the clerk will talk to the customer here is a special deal for you will be interested to bang this one or bang other things and it turned out to be very effective and of course this can be used kind of in online shopping as well during the two months trial period after we develop the solution using both the personalized the recommendation as well as the the associative analytic the we found out the the recommendation in people call a click-through actually is 10.1% of the recommendation get activated means get the transaction and compared to other store which are not doing this experiment I mean the body shop has many store we only select a few to do this experiment and it showed up in two months the income for those store using solution increased by 6.8% which is really good and even a single this body thing is is it's good okay the second use case example is in smart manufacture so you probably heard the term industrial or industrial 4.0 or productivity 4.0 in either case I think the the things is related to manufacture a Mackenzie did a study showing that the manufacturer sector actually store more data than any other sector and in even in 2010 it already store like close to 2x up by of data and and given so much data there's opportunity for and analyze it and to get the the business value and what one can get is you can try to increase the business value by lower down the schedule unscheduled downtime reduce men and of course improved overall equipment efficiency or improve the return on investment so we have worked with three IC manufacturer company in Taiwan and what they want essentially try to find way to improve overall if equipment efficiency and for a specific IC company we work with they have tried to do several of things here but mostly manually through expressly or through certain manual inspection and what they like to do is to able to automate them or at least semi-automated the process to get all this thing resolved or to have a solution for this okay and to do that we came up or we work with them to develop a few of the analytics solution including anomaly detection and core our root cause analysis predictive maintenance and optimize the the schedule or job page I will not get into the detail here I just show you the achievement so after working with them after we kind of collect hundreds of log information or look from hundreds of equipment are keeping in mind each equipment has many many look many controller many sensor generate so it's not just hundreds of there but if you count all of this is south thousands or maybe hundreds of thousands of them and then we process and extract information from billions of log and also other data set and by extracting combined analyzing this we generate some prediction and analysis and build a certain model for prediction and we validate those model and the result and it can achieve about 90% accuracy this is compared to they already doing manually through some other thing and we compare the result and accuracy is about 90% and this is even more important because the way they are doing now they cannot do anything real-time response and with applying the big data analytics using the the swimming technology like storm we are able to have a real-time response and help them to a minute to do inside delivery and also to have a fast response for example in the pipeline or manufacturing delay of one minute or two minutes may cause a lot because if something is run in high-tech manufacture is run if you don't detect it whatever in the next two minutes whatever product you are generally is it's going to defect it is useless and and this they are very appreciative that we can help them to achieve this okay I don't have time to go through this bonus we also apply in the health care so in summary big data applying big data analytics solution to improve the operational efficiency has caught every business executives attention and the problem is how to help them or enable them to be able to apply this more efficiently or more effectively and that's why we in Taiwan we build a big data analytics platform for open-stake which has emerged as a leading technology for building the cloud computing platform and deploying cloud services so we combine the two we build a BISO SE and we have shown a several big data analytics success story in Taiwan and all I mean we have explored open-stake to deliver the cloud-based analytics solution and finally it's think big think open and of course please think BISO thank you any any question maybe time for one question if not okay thank you for coming again thank you