 So a very warm welcome to the last session for today Later on we come to the fun part a lot of good stuff at the booths So I hope you will enjoy also some relaxing drinks at the exhibition after my presentation in This session I would like to cover the topic of SAF Which is a fascinating technology to build the future generation of storage as we all know The cloud is changing everything. So it's also changing storage in a sense that It's getting unpredictable What type of capacity will be needed when going in a broad cloud project? Today, we have a very stable Data growth every year, but the cloud is changing everything Yeah, if you are a cloud provider or you start to build your internal cloud You never really know which type of workload will probably brought on the storage system when and in watch quantity Now if you take a normal storage architecture of today, typical scale-up rate systems They are very mature. They're very reliable They have a certain scalability But they have a shortcoming when they're coming to the cloud and the data are rising and there's a kind of Divide I call it the petabyte bit divide I believe there is some lots of free space and Front rows. I'm not spitting But there are some shortcomings, yeah So if you scale beyond the petabyte a high-end storage system is very very expensive And they will probably have scalability limits in terms of capacity and performance Another big issue is the rate Construction rate is a very efficient way to care for data redundancy But today we have four terabyte disk if a four terabyte disk fails The rebuild time could be one day two day if it's a fully loaded systems even longer Now imagine the future six terabytes are coming eight will come and ten have been announced So the probability that another disk will fail during the rebuild time Of a disk is rising. So data loss gets much more probable than it used to be Then of course data migration think about you have reached the petabyte divide or hundreds of terabyte And you have to move them after some years to a new storage system That takes a pretty long time is a little in a huge effort And then of course cost per gigabyte. Yeah, we all under a cost pressure There's nothing for free in IT anymore this density data sender space. So it's about Think about new generation of storage infogy So we were as a storage vendor we had a project running now for two years and it's now Finalized and the product is out. We were thinking how could the future of storage look like first We made a requirement list What is actually needed so kind of a Christmas list Christmas list if I can wish something first of all in the cloud I need modular scalability not only from capacity, but also from performance. I need zero downtime You cannot recover petabytes of data or in the future petabytes of data Easily, yeah, so it should always be running it should be Possible to refresh the technology if you have tons of data it should be capable of Refreshing the hardware once it's end of life, but still keep the system going. This is why Everybody is looking at software defined storage because it's separating the intelligence of the system from the hardware And you can replace components and the storage system can theoretically live forever As long as the storage is a way a software is available data management, yeah If you're in storage, you know how hard it is to tune a system and it will get worse when you have tons of data So it should be somehow self-optimizing so automating the service levels to the end users It should have an efficient management to deal with this petabytes of data and Of course, it would be nice if the total cost for the capacity would go down and not go up then Yeah, today what's quite common is a block device and file devices But if you increase the number of files today's file systems will be at its end So we have already customers with billions of files and traditional file systems supported anymore So its object is on the rise object storage will be in the rise and in order to avoid to have different storage silos again like in the past Also in the present such a system should cover object file and Block support Now when evaluating what can be a future technology two years ago, we started to evaluate self Whether it's a software which is also for larger implementations and we were surprised how stable it is running in its core What you can do theoretically you can just put stuff on a cluster of servers and build a very very scalable storage system it's really a Fascinating how this technology has involved and the good thing is it's hundred percent compatible with open stack So if you're on a cloud project, Jeff is a good choice for a storage software to implement it in the cloud environment Now when we were evaluating Seth and we were thinking of a big installation We immediately found out that it's a hard job. First of all installing Seth on a server is a Very easy deal and connecting service is also not difficult But think of failure situations a note in this cluster fails How fast must be the interconnect be between different storage nodes to recover the redundancy again? How do you maintain the whole stack? Yeah, so We did it also in our labs had some old servers because we don't had to go for another investment round So we had a mix of servers connected and it was performing like a big storage system But maintaining it was awfully difficult because every new version of Seth every new patch of Seth Was influencing the drivers so it didn't fit with one storage a server system It didn't fit together with a network component. So maintaining such a stack is extremely difficult So it's kind of a research project internally Then the maintenance part was really awful. Yeah, of course, Seth delivers a lot of main management tools, I mean But of course you have other tools to manage the server to manage the network to manage the drivers and so on so it's Instead of a normal storage traditional storage system You have typically one management console and here now you end up with ten or so. Yeah, because it's a stack Then of course if you look also on a commercial point of view You have now different components from different sources Yeah, the the server from one vendor the connectivity from another the Seth from whatever distribution you have chosen And so on and so so in a in a nutshell you have to deal with Five six vendors for this storage stack So it's also from a commercial point of view Managing the maintenance contract a difficult thing and then of course, it's a troubleshooting in a normal traditional Rate system you can go to one vendor if there's a glitch say please fix it I have a maintenance contract, but if you build it of your own Then this storage system is existing only once in the world Yeah, if you go to a traditional vendor by a normal storage system this system will be a hundred or thousand time out And there is a standardized support, but here you have built your own storage system And you are now your own to do the troubleshooting and then the safe guys say no no It's a networking problem the networking guys you say, you know, it's a server problem somebody says is an operating system problem and so on and so on and This is not real fun when you deal with petabytes or terabytes of data. Yeah so Instead of doing this with we're thinking what can we do as a vendor to make it easier and this is why we said Why not delivering it out of the box and this is what we Recently released and you can see although in the booth After the session is a system we call eternal CD 10,000 the C stands for cloud and it brings everything together the SAF software The compute hardware the storage devices the interconnect and the maintenance so it's the one appliance Which can scale to a huge extent and you have one vendor you can go to in order to solve any problem So this is how it looked like at the bottom you see a cluster of service or we call it storage nodes There are different flavors basic storage node for development purposes Capacity nodes with lower spinning disks to host a lot of capacity and performance nodes with give a very high Oh and they have SSD caches and so on and so forth and they are connecting With a very fast interconnect. That's a infini band. I tell you later. Why and to the front And it's standard ethernet 10 gigabyte ethernet the brain of it is SAF. We have chosen the ink tank and now redhead distribution and We have added a management layer on top Which gives you one single point of control of every component in the hardware And every component in SAF so you have only one management system to run the whole Construction and this is also important. Otherwise you have to deal with five or more management tools Then of course we are contributing to the SAF community. We found several glitches for High performance environments where we need a fast storage. So We are already contributing to the core development in order to fix some glitches At the beginning we will support object level access But also block level access so can you can use it also as a standard storage system? Even if you don't in a cloud project file will be released at a later point of view This is something in the roadmap in 2015 Because SAF related file system is not the brightest one on the market This will be coming later and wrap wrapped around is a complete management Service so whenever you take an eternal CD 10,000 You call one vendor whenever there's a glitch in the core in the SAF in the interconnect So it's like dealing with a normal storage vendor Now what is actually the advantages of such a system? By just adding nodes you can scale more or less endlessly. I show a little bit later how far it goes The construction is made in a way that there is no reason to have any downtime be it in a failure situation or in a maintenance situation It can be refreshed online. So you can pull in new technology without Driving the system down. So it's more or less immortal It will support block file and object And it is also cost optimized because it's built on standard components And it's using standards in this case SAF, which is a perfect fit to open stacked projects Now, let me go a little bit more in detail It was double One more up So in the system Wait a minute. Okay in the system the SAF algorithm is responsible to bring the data in a kind of Intelligent distribution to the different nodes and different discs This is also a big advantage in a sense that All the a file for instance is highly distributed in terms of data blocks within the system Those there's lots of parallel reads and writes if you are using the system. So even with slower disk, it's gets very very fast And it has a kind of a built-in High reliability. It's not a rate system anymore. So the redundancy is you define in one volume I have two replicas. It means every data block is there Physically two times if you see I have higher service level and I'm more careful Then you can decide to have three replicas for five if you say I want to save money There is something like called like erasure code. Then you have a one point three relationship Which is make it marry efficient. So Some people already said yeah rate is more efficient. So it's not too much spare Capacity used It's depending on your service level and you can define it by volume. So you can give different service levels To the end users. So now in case a complete note fails Probably with lots of disk This mechanism cares that the redundancy is done or rebuilt in some instances because Many discs are now writing on many discs So it goes very fast. And this is also why we have chosen in Fini band as a back-end Infrastructure that a note failure can be recovered or the redundancy can be recovered very quickly This was also a learning about two years of testing and developing If you are not caring about the whole system architecture, then failure situations can be very harmful Now I say immortal system. Okay. This is a very bold marketing Statement but in a traditional rate system There is a certain point in time either the system is full and you have to buy a bigger system Or it's so old that you can't get any spare hardware anymore Yeah, and then you have typically have to do a data migration and if it's a very big system Then it can take a very long time What you can do here is just add a note and once the system has detected the note It redistributes data in an intelligent way and you have more capacity now if this Hardware what you are using is no longer longer available then you add the next generation of notes and The system keeps on running and then all after time you can replace the older notes with newer notes Always online. So it's like an organism with some cells Yeah, if some cells on your skin are dying, they are replaced by fresher younger cells And so you can keep the system alive As long as a software in this case safe is existing So probably safe will not exist forever. There will be in 10 15 years some bright new inventions, but For the foreseeable time you can create a very very long lifetime of a system So in fact, you don't have to migrate for a pretty long time now block file and Object access already mentioned file coming a little bit later. It's on the roadmap in 2015 This is also important in order to avoid Newer storage silos because here here for highly scalable systems So why not having every type of data on one system and it's not a good idea to create new silos again so both there may be also a future formats of Storage access so block file an object is just the beginning We expect there will be in more variety in the future for instance a lot of this virtualization guy are Constructing new types of storage access like VMware is now short-lived for releasing we will and such type of access methods So there will be more specialized methodologies in the future, which can be built in such a system now cost optimized first of all Not automatically a software defined storage system is cheaper than a conventional system because as long as you are using enterprise quality disk The total cost of order total purchase price of a system is defined by the number of discs But on the other side there are some things which will bring down the cost first of all Such a system is based on standard x86 hardware. So it needs no special development So the total cost can be and the efficiency the scale of its efficiency is better So the cost can go down Then at the open soft software opens open source of that's F is of course Bringing the cost down Especially in a sense that a lot of functions where you normally would pay extra in a classical rate system for high availability for thin provisioning for Kind of replication mechanisms are these are all extra costs typically priced by volume This is no longer necessary because it's part of the design. It's part of safe Yeah, so it's one software with all-inclusive Seth delivers also a lot of automation feature how to rebuild the system how to rebalance the system how to replicate data Yeah, this is also part of the initial cost. So there's no extra Bringing costs down. So it's affecting more the total cost of storage if you just Would take the raw price per capacity This would probably not make a much difference But if you estimate the total cost of ownership, then it's a completely different picture then of course the The longer on a longer run you have less migration costs Because the system is just refreshed or expanded as you need it. So there They Storage migration project is a costly thing or at least cost a lot of effort if you do it along The long life cycle is also paying back. So let's complete new investments are necessary and Of course as we bring it as an end-to-end support you save also support cost you you get it in one shot instead of having diluted kind of a cost Distribution by many vendors So delivering many services for many components So what is actually then different instead of building a safe base storage of your own first of all Although it's basing on open source. It's a maintenance storage system Yeah, so you don't have to care about it So and if you are on your own, then you probably will face a lot of trouble shooting The end-to-end maintenance are already mentioned Adding new functionality is also an hour roadmap where at safe has gaps. For instance, we will introduce the ember driver We have embedded a lot of SNTP traps to have remote management. So this is all these Extras a normal safe environment Cannot deliver today Then of course integrated management of all components. This is also something you don't get it from the shelf It's a performance optimized sizing So we were testing a lot how to build a cluster environment and to standardize the cluster nodes in order to make them reliable and also very performant in a daily operation and Then also very interesting. We are starting to add apps on top of the system Which can help you to provide intelligent data related services Let me give you an example a real-life example of a one of our customers in Germany, there's currently a better face of a very interesting project That's from the German stock exchange They have built a daughter company which now is using a trading platform for trading IT resources If you go to this platform and it's live if you go to the internet link You see what you can purchase or let's better say rent for a certain time. You can rent CPUs RAM and this capacity and so a lot of and the whole platform is based on OpenStack and Now they have already first customers consuming these resources But of course as this company does not provide resources itself. It is it's relying on partners who are delivering This compute and storage resources. So now put yourself in your shoes You are now a service provider, which is here seller one seller two How much storage which would you purchase? Yeah, you don't have any glue because you don't know How much users will enter this trading place and consume how much storage so for a provider in such a market It's extremely difficult This is why a data sender Decided for this eternal CD 10,000 system because they start very small. They started with roughly 70 terabyte Which are they are offering in this marketplace? But if this marketplace takes finally off Then they can scale More or less endlessly and they have no downtime and so they can offer very good service levels at an interesting price to this marketplace I Said we will bring apps on top of the system so Currently we have added An application called irots irots is an open source data management system Which helps you to build large archives. So it's about data management and retrieval We are thinking about adding sync and share software on top of it to build a kind of a drop-up type of environment So we are currently approaching a lot of ISVs and software vendors Trying to bring them into this ecosystem. I give you now an example of a real-life example from another customer a University in Germany that's in a small city called Mainz They have a new university of roughly 15,000 students and several hundred Professors They had Distributed IT so each department chemistry computer science physics They had their own IT in their own storage and they said no we can't go on like this We have to offer a central IT and with this they wanted to have a centralized storage and they are Oblige by law and by some regulations to keep all the data ten years online all the doctor Thesis is all the test data of experiments in the physics and the chemistry department. Yeah, and of course But 15,000 students in ten years time produce a lot of Stuff yeah, so they had also the problem how How much capacity should we buy and should we buy a high-end system? We cannot afford it. We are a university So typically universities are not rich so they say is decided for an eternal CD 10,000 saying we start small and We are adding more data after time and they put irons on top the irons software is for their Library services and for the retrieval of a lot of tons of data the next step is now that they want to implement a safe sing-and-share Service for the university campus and they want to differentiate which data can be publicly accessed and which not and this is why they are now in the next project Deploying a sing-and-share Software on top of the system Typically use its areas. So what a software defined storage system and this type is good if you have Big data amounts and if you have uncertain growth if you are in the area of a couple of terabytes then It can't be recommended then it's the best thing you go for a standard rate system because a scale-out system Needs a certain basic investment because you have By by by design you need a couple of nodes. That's a system that can really work And if you have only few data volumes, then it doesn't make sense. So Sensible thing to start with is if you are in the area of two hundred Terabytes and more in production. Yeah, of course for test reasons. They are typically smaller now typical usage areas are Open-stack Environments because it's based on self. So it's completely integrated from the scratch If you are any type of service provider being internal service provider or external service provider for the cloud Because then you have this uncertainty How much data you need tomorrow? Telecom businesses we are so we are currently talking to some of the telecoms who are now getting IT service providers Offing offering syngen chair or some data services on top of the typical telecom business University on research organizations. I mentioned before But also in the public sector. This is very interesting because they the public survey sector has tons of files about the sentences text or police or whatever so Having tons of data online Then when you deal with Multimedia because it's a highly distributed system So there are only small data chunks on one disk and they're highly distributed So the read mechanism is highly paralyzed writing mechanisms are highly paralyzed or ideally for streaming and writing media Business analytics in a sense when you have to aggregate a lot of historical data and keep them online for analytical reasons and We we currently have also contact with a lot of financial institutions not for replacing their core storage system This seems to be go on for a pretty long time with standard rate systems But the cloud projects of financial institutions here. We have some proof of concept Now the in a nutshell This is set out of the box without the hassle of a lot of experimenting and Having hassles in maintenance. So it provides more or less unlimited scalability In its first release, it's covering 56 tera petabyte So should be enough with more than 200 nodes. We haven't yet seen a case where somebody has so many data Perhaps the NSA would be a good example, but not a customer of us It has zero downtime by design. It can live forever at least as long self exists Unified in its completeness cost optimized by billing our standard and we are working heavily together with redhead and ink tank Constantly improving the system if you want more know more about the system you find us on the booze C4 This is in front of the auditorium part Actually as Fujitsu is a Japanese company I want to highlight that today a magazine voted that a Japanese whiskey brand for the best in the world and We have it on our booze so you can Test it and perhaps also get some spirits about self on the booze Thank you for listening. If you have some questions, please shoot now. Thank you Please Excuse me yes, the control panel can do magic things you can Say I build pools I can distribute data to certain locations And such type of things so we can we have a demo on the booze and you can see how it's done Whether we have control about the crush algorithm, yeah, so if no questions right now There's the opportunity still today and tomorrow to visit after the booth. Thank you for coming