 I'll jump straight to the talk. This is a collaborative work with other researchers from the San Diego Supercomputer Center and EL University as well as our last grant is jointly with the Silver Lab at University College London. So these are the three sections of my talk, except I may not make it to the last one, because just for the time-time constraint, I will see how fast I speak. So I think these are, I will go through these some slides very fast because there's nothing to say here to this community, you know, the US Brain Initiative started in 2000, starting in a revolutionalized understanding of the brain, funding from a lot of agencies in the US, NIH, NSF, Department of Defense, private organizations, and create a dynamic picture of the brain, and then, you know, try to treat cure and even prevent neurological disorders. So that's the US Brain Initiative, and then at the same time in the EU Human Brain Project started, they are taking a more simulation-oriented approach, you know, simulate the brain in 10 years, and map all of the neurons and synapses, and again, you know, eventually going after in understanding neurological diseases, and simulation and data sharing and the collaborative environment are a big part of the human brain project, and they release their collaborative environment this March. So now, talking a little bit about the large-scale simulation, I come from the San Diego Supercomputer Center, where we are always excited about large-scale simulations, and again, a lot of you know this. If you look at the groups on the left column, and in the years, you know, there are simulations of 100 million cells going up to, you know, many hundreds of million cells, and billions of cells using 100,000 processors on different kinds of supercomputers, and using the K-computer in Japan, this Disman Group in 2014 and 15 did 1.86 billion neurons, but as we know eventually, you know, there will be extra-scale supercomputers available in one of the countries first, and then followed by other countries, whether it's in China or Japan or Europe or USA, you know, there is a very clear target for the high-performance community, which is always good to have a target, and there is some competition among the vendors and countries and research labs. So, you know, when I talk to the neuroscientists, computation neuroscientists, they also, you know, feel excited about having accessible, to get their hands on accessible computer. And again, into this community, you know, data processing is obviously a big part of the neuroscience community, and it's getting more and more mainstream, and a lot of, you know, the previous talk about data sharing. And we are a little bit focusing on data processing, and when you want to process large amount of data, if you are, you know, accessing a supercomputer for processing data, they're the issue of data transfer, data storage, and again, the data sharing, all these come into play. We are a little bit focusing on what can we provide for the broader community, not the extreme scale, and I know some research groups in the U.S. I'm sure this is true in Asia and Europe, too. You know, they're following, like this clinical example, thousands of individuals over decades, and they then have need for hundreds of petabytes of storage and data sharing. So those are exclusive people. They will have their funding. I'm trying to see what we can provide cyber infrastructure to the broad mass of neuroscientists, not the elite ones. Elite ones will always have funding. What can we do for the broader community of the neuroscientists? So now the motivation for the neuroscience gateway is, and I will explain what a science gateway is, and I'm sure you already know it, is basically providing a web portal for a specific scientific community and try to hide all the issues that you have to deal with running on a supercomputer. Some of them are administrative. Some of them are technical barriers. Just hide all of those from the, whatever the community it is. In this case, it's the neuroscience community, so they can use supercomputer. So motivation for the neuroscience gateway was, as we again know, there has been a tremendous advance in computational neuroscience over the last two, three decades. New journals came out that are focusing just on computational neuroscience and modeling. The funding agencies saw proposals which are, even if they're from experimentalists, they had modeling as a big part of their research, or they're purely modeling simulation-related research. And at the same time, there were a lot of development of tools, you know, the neuron, M-cell, genesis, mouse, free-surfer, FSL, all of these tools, both for computational neural network development as well as for data processing. They're open source. They're meant for parallel computing, both for simulation as well as data processing. And at the same time, there has been tremendous advance in cyber infrastructure, of course, you know, going from hundreds of teraflop machines to hundreds of petaflops, and maybe in six to seven a year, depending on which country or five years, you'll have an excess-scale computer as well as data analytics. So all of those have been happening over the last, you know, decades. And so the neuroscientists, many of the, I'm talking a broad mass of neuroscientists, graduate students in every university, for example, who are doing a simulation of data processing. They start out small, but then they might be forced to keep their research small because they don't have easy access to high-performance computing resources or cyber infrastructure, developing data transfer storage sharing. So, you know, like the complex network modeling and a lot of network optimization and data processing. So not everyone has access to HPC resources easily. So that's the point. And then although there are national academic supercomputer centers in every country, which is free, but you still have to write a proposal every year to get time on those machines. And that's very competitive because you're competing with astrophysicists, biochemists, earthquake modeling, climate simulation, molecular dynamics, maybe social scientists. Everybody's competing for time on those academic supercomputer centers. You have to write competitive proposals. The time available is always less. In the U.S., usually it's two-third of the time that's made available versus what is requested. And then you have to understand the HPC machine and storage and the Q-system and the authentication and data transfer. So these are all technical administrative barriers for accessing HPC resources, although in theory they are available. So now we'll talk about the neuroscience gateway. So the idea is very simple. We just hide all of those, hide or lower the barrier, administrative and technical barriers so that as a researcher, you just get an account on the neuroscience gateway. You upload your model or data, and we provide lots of broadly used codes in the backend on the HPC resources. You run your upload your model, select some parameter related to the code or the data processing code, submit your jobs, then later time when the job is done, because usually in supercomputers there are some Q-wait time. You get a message, you download your result. So basically hiding all the administrative technical barriers and making computation neuroscience, democratizing computation neuroscience for a broad mass of neuroscientists. So that's the whole idea of the neuroscience gateway. It was initially funded by the U.S. National Science Foundation starting 2012, and I'll talk a little bit more about our second grant. So just to show you the number of users that has grown over the years, we started, as I said, in 2013. Early we had about 100 or so users, then 200, and now we're about close to 400 users. I should mention that at the beginning, say 2012 or 13, when we started talking about the neuroscience gateway to the community, initially people from all kind of people got excited to get an account and see what it is about. In the later part, so not all of them ended up using, because it's not exactly what they were thinking. Maybe they didn't need it, but it just got an account thinking it could be of use and need. But later part, starting later, 2014, 15, 16, these are the people who exactly know what they need HPC for. They already had a model. They had a plan to do their research, and suddenly there's the scope of doing the research in a bigger way. For example, one postdoc I remember, we know he graduated now, he started a postdoc around 2013 or 14. I think he had planned out everything he will do in two years for simulation. In NSG and neuroscience gateway, he was able to do it in two months, so he has to completely have to rethink his whole postdoc plan in a good way. So that's an example of an impact. So now I want to tell you how many supercomputer core hours we are providing since we started. In 2013, it was about, as you can see, 187,014, about 600,000, 2015, close to 2 million. This year, close to 6 million hours, and it was used up by the users of the neuroscience gateway. So again, I want to say that in the back end, we, I and my colleagues, we do the laborious job of writing an allocation proposal submitting to the U.S. Academic Peer Review National Supercomputer Review Committee, which reviews our proposal against the astronomers and earthquake scientists and molecular dynamics everybody and then gives us time. We justify based on how much it was used and how many users we see, we will see the growth and all these things. I have to say they are very supportive because they see we are providing a service to the community, so they are very supportive of that and you can see how the core hours are increasing and pretty soon I will write the proposal for the next calendar year, I think I'll end up asking close to 10 million hours and every year, during towards the end of the year, our time runs out and I have to go to the committee and say, please give me a minute to read because they realize we are serving a broader community. So here is a since 2013 to middle of 2016, I just wanted to show you the maximum core count job that was run as Supercomputer on each month and the trend line. You can see the trend line is showing it should be 2,000 cores or more, maybe a little more than 2,000 cores now, but over the years users have run 4,000 cores in simulations. So that's one trend line and this one is again the maximum or the average core count job since 2013 and I think the overall average over the year will be few hundred cores and average they are running on few hundred cores on the Supercomputers and on the back end we provide access to Supercomputers not only at the our center, the San Diego Supercomputer Center was also at the Texas Advanced Computing Center and at the Supercomputer Center, so back end we use different kind of Supercomputer which is good because each has his own strength, different processors, GPUs and all these things. So just and as I said we have close to 370 users, not all of them are active at the same time, maybe it's 20% of them are continuously running depending on the research what's going on with the Grass Student Life or Postdoc Life and these are just some examples of different publications or research going on like the group from this is the Litten group from Sunni Downstate they are doing some evolution algorithm for developing neuroprosthesis and then another group Team Rumble, they did you know ion channel conductance and kinetics models of Reusers Monkeys and then so all kind of research going on by different groups and there's Alan Institute this is Catelyn, he's a graduate student working there, he's working with other research at Alan Institute, they are doing trying to develop a ground truth for spiking model, they are producing tens and tens of terabytes, those are good headaches to have for us to see the science progressing but if everybody produce tens of terabyte 370 users we'll be in trouble so it's interesting perspective we see by running the neuroscience gateway since late 2012 until now I can see kind of as observer of how the neuroscience field is or how the simulation needs are changing going from computational to more data processing people so it's very interesting to see and also see the issues that come with it, the data sharing and all these things. Here examples of other things going with the neuroscience gateway the radiation oncology group at UCSD they are you know they see that radiation treatment for tumors can result in neurocognitive deficit in neurocognitive function down the road so they are using free surfer understanding brain segmentation so they can track those things and see how it can better improve the radiation therapy the human brain project this is Michel McGleary from the human brain project has been working with us his plan is to release a lot of the benchmark and models and tests which through the neuroscience gateway to the broader neuroscience community and EEG lab group at UCSD is going to make a multimodal imaging software improvement of the EEG lab software for processing they want to make it available to neuroscience gateway a lot of things going on I can see researchers developing network modeling specifically focusing on GPU because these research groups believe GPUs are the way to go again they come to us because we can provide access to GPU HPC resources Genelia Farm is looking into analyzing scalable image using some latest park processing environment so all of this we can provide at the back end so it's interesting for us to see how different you know we started with the large network modeling but now we can see more and more data processing people are coming to us which is interesting observation so far what I talked is the neuroscience gateway you access it through a portal you just get an account and you know do your things using the course we provided the back end but then last year NSF gave us a second grant to develop a programative access to the neuroscience gateway so instead of a portal you can programmatically access the HPC resources of the neuroscience gateway and that is jointly at the same time the BBSRC in UK funded the silver lab in University College London who are the developer of the open source brain so open source brain will be the first community project that will programmatically access the neuroscience gateway HPC resources the reasoning behind the programative access was there a lot of big community like open source brain and NIF and ModelDB who has a large community of users and we didn't want them to all individually get an account on the portal and use they can from their familiar community project environment directly use the neuroscience gateway so that's what we are working on so open source brain again I think I'm sure a lot of people here know here it's a model development repository from the silver lab at the University College London we are working with his research and his party Gleason and so models can be visualized and converted into a new model to other network tools and so the idea is from the open source brain directly simulations open source brain users will be run on the neuroscience gateway using the restful web interface we are providing and again this is you can see various kind of workflows can be accommodated this little lab is developing large scale network modeling simulation is a net point so you can upload your model and open source brain modify parameters create a net point model send it programmatically neuroscience gateway get the results back visualize the result all these kind of workflows can be accommodated through the programative access of the neuroscience gateway let's see how much time do you think I have another few minutes just to summarize the neuroscience gateway as I already said in the various ways the neuroscience computational and data processing neuroscience community any user from anywhere in the world can get account on the neuroscience gateway no restriction at all and if US researchers you know so the neuroscience gateway is not for providing someone 10 million hours it's for the few hundred thousand it's for the broad masses if someone wants to get 10 million hours in the US then they have to write their own proposal and get allocation and compete with everyone else so I think we are doing a service for the neuroscience community by providing ED access to HPC since I have few more minutes I will just briefly mention what we are doing next we are trying to understand the evolving overall cyber infrastructure need of the neuroscience community not just HPC we are again we notice how the neuroscience communities processing needs changed over time as we ran the neuroscience gateway we are running a survey with 20 questions and almost 100 people have responded from all branches of neuroscience and I will just show few quick ones just to be able to get an idea of what is the size of your research lab almost whatever 80-90% is between you know 5 or 5 to 15 and then status of the responded most of them are grad students and faculty postdocs and then what are the research data type obviously experimental and imaging and computational data type seems to be the most what is the size of your current research data if you summarize 65% within a terabyte, 35% 10 to 100 terabyte and the future growth rate of data again if you can look at the very for the broad masses you know few terabytes per year and how long you want to store your raw data or your final data again everybody you know depending on which responded most of the ones to store it for 5 years or longer you know you share of data sharing you know you share with immediate collaborators or collaborators internationally and then limitations for example most of them say I want a first restricted access for maybe 1-2 years either finishing my PhD or publishing my results in the journal then I will share it with the community and how do you want to share the data you know they want to sync data with the cloud and the HPC resources and then again it's mostly you know few thousand cores to 50 thousand cores but it was interesting to see that at 13% things they need access scale computing and obviously 90% of them says that a collaborative environment is very important for them so that's what we are looking into going forward I'm interested by your work on the rest APIs I think quite a lot of similar platforms around the world now and I think it would be very useful if we could coordinate around common API I'm thinking about Agave in the US Seabrain in Canada and other platforms around the world and I think it would be very useful for interoperability these all these platforms could expose the same API and since you are working on it can you talk about it sure that's all the reason we came up with this thing is to establish more collaboration and know what others are doing so we can do things jointly learn from each other we have to move on thank you very much