Lab to Live: How Pure Storage accelerated Folding@home’s quest for Covid-19 therapies

And my name is Brian Carpenter of senior director, unstructured, technology strategy for Pure Storage and I have with me to very important people to talk about, not only what we're doing from a technology perspective. But how do you know, how how are the teams at folding at home? Really are working on that covid-19 problems. So Stacy and Greg want us to go and introduce yourself. Yeah, Stacy Brown. I'm a field. Solutions. Architect here. If your storage I'll pass it over. You Greg will medicine in St. Louis.

What are the reasons why we're here today? Greg is through a series of very fortunate events, some industry partners of yours and ours for have introduced us and in the very early portions of the kind of covid-19 pandemic. It was while I was over a year ago. So it's not only we get to learn about what you're doing and fully get home has been around long before the covid-19 pandemic and all the things were going to do here. But also the things that you did

through the last year are obviously very impressive. So that's why we want to talk today. And you know, what is it is a great opportunity. Greg. Did you have anything to say real quick? No, this is amazing. I I we've been having a lot of fun working with you. So appreciate it. Thank you. And see you tell everybody. You know, I don't have your storage installed. So why don't we talk about what we do in relationship to what folding home. And we're helping folding

at home, access that data to turn it into valuable information. So part of the computation is a course leveraging. Every person's home computer with the whole thing at home binaries. And then that information gets to take him back and stored on a flashblade, which is our product that offers object and NFS storage and then that enables folding at home to access that data and do really creative useful things to help find out what's going on with covid and how we can help come back covid. Those are pretty cool. I mean getting to wake up in the morning talking about how we help people

with a i and you know, computational science and all these other kind of emerging Technologies. I was just on the phone talking about genomic sequencing which right now is better than both of us. But, you know, like in at the end of the day, being able to talk to people about things like ransomware and all those other really interesting things. It's a really fun day to come to work, but it's not as fun. Let's be realistic. It's not as cool as what Greg got to talk to you. So why don't we, why don't we ask break some questions instead of talking about herself or

not? A new company is Brian touched on it earlier. He hasn't been around for 20, 20 plus years as a company, and we've heard, congratulations are definitely in order for you. As you are going to be in the Guinness book of world record for the fastest supercomputer. Can you tell if it's a little bit more about folding at home this mission? Yeah, it's been a quite a exciting ride over the past 20 years and things have evolved a lot too. But the currently we see our mission is really empowering

anyone with a computer and an internet connection to become a citizen scientist and help us combat Global Health threats. Threats are are wide-ranging. We've focused a lot on you know, what, I'll call Tisha Mobile Medical Health threat. So Alzheimer's disease Ebola, virus cancer. Quite a lot of the antibiotic resistance are quite a lot of alarming, rest of all of our. Our health started branching out into, you know, all those things that's are of a similar magnitude. Like you have increasingly the efficiency and productivity

of crops, for example, and dealing with pesticide resistance and starting to think about energy and climate change and how we could play a role. They are. So it's the Oh really funny area where we get to be at the Nexus of a lot of different things from a i to physics and chemistry and biology and Computing and you know drug Discovery and genetics and and all of these different areas. Of course, Greg, you mentioned all these things that you've already been doing and it really focuses from my

understanding around simulating, protein, Dynamics and those types of things. But obviously, this past me notebook, 12 to 18, months, whatever it is. I've been really different. So what kind of dig into what you really been doing, it is in and how you're helping fight the baby. Yeah, so are very general interest as you said is in protein Dynamics and your for those that to your proteins in our thinking about, you know, what meat will be in your entree this evening or your protein shake. Ya really. These are at the molecular machines that are responsible for most of the active

processes. We associate with life, right? So it's proteins that allow you to control your muscles proteins that are detecting light in your eyes proteins that are sensing vibrations and in your ear so that you can hear sound, you know, it's proteins that are responsible for free. Turn down food, you eat, then the building other bits of your your body. So they're quite busy and and varied in their functions. And when they malfunction, we end up with things like Alzheimer's disease or cancer and proteins are used throughout biology. So when we think about

infectious diseases like bacteria or viruses, a lot of what they're doing is based on these molecular machines, and one of the challenges is that you like the machines that were used to thinking about on macroscopic scales, like, our cars and trucks and whatnot. These things have lots of moving Parts. But because they're in such tiny like scales near nanometers in lower. We can't zoom in on them with any existing microscope, microscope, wash them in motion. We just get the single snapshots of

what they usually look like, right? And so if you think about your car, For example, you know, your car is quite interesting. You can do j-turns and donuts and get to work. But lots of times, it's not doing anything interesting. It's just sitting in the parking lot. So if you had a snapshot of what your car normally looks like, it's just a big piece of metal parked in the parking lot. And if you didn't know what it was, it doesn't seem all that interesting. Right? And so what we do with folding at home as we take these individuals, snap shots that you contain a wealth of information but far

from complete. And we use computers to simulate all of their moving Parts in and really get into the details of how they work and what's going wrong, they fail to work and this is a tremendous value. If we want to use a redesign of protein to do something else or control, if they with a drug, for example, because they are just as it's hard for me to think about understanding how my car works and how to fix it or how to make it go faster without seeing all of the moving Parts. Brenda the same situation. So well, what happened at the beginning of 2020. When I started

becoming clear, that covid-19 was going to become a pandemic, is that we never put all of our varied efforts in Alzheimer's cancer, and antibiotic resistance with these are tremendously important but there's something new and more important right now and put all of our intellectual one copy tationil resources into understanding how the protein components that allow the Stars, Kobe to virus to infect our cells and replicates and spread at Walla dating. Our immune systems, how all of this works so that we could help inform the development

vaccines and Therapeutics. Yeah, so that sounds to me Greg. You know, anytime you think about self-driving cars and the amount of data that you're you're taking similarly in all of this trying to stimulate that what's actually happening with the proteins and the covid-19 virus. That sounds to me like it's a tremendous amount of data that elected. How did can you talk a little bit more about some of the challenges that you guys ran into when you were collecting all this data and trying to make sense of it and

the ability really for you to Crunch that data fast enough. Obviously. Everybody was anxious to in the pandemic so we could retrieve information. The better. Naruto is a very interesting time both scientifically and you know personally and and technically the beginning of the the pandemic because of course, you know, we were all dealing with the same things that everyone around the globe was was dealing with you personally and professionally, but then with the new folding a home, we were an interesting position where we've been operating for 20 years and we

have 30,000 active devices around the world who are helping us to run these big calculations intractable by any other means. And so we were regular running these massive calculations and generating your data sets that even in there yet strip down and compressed formwork, you know, on the order of tens to hundreds of Gigi, be happy to share this data with people. But often at that scale, you know, it's hard for us to just post on our our website so we would end up. Yeah, literally, mailing hard drives around because that was the most efficient way. To

deal with that the scale of data. And so what happened at the beginning of the pandemic? When we first shifted, our attention and launched our first simulations of proteins from the coronavirus on full home is that your people were all hunkering down at home and you're scared of the future and you're feeling pretty helpless than like, they couldn't do much of the Hide and we were able to come and offer people a legitimate opportunity to take Parts in going on the offensive against this new threat and trying to figure out how does this thing work? And how do we how do we help stop? It

said within the first couple of months of the pandemic we went from having 30,000 doctor's advice is helping us run simulations to well over a million devices helping us run simulations. And so, you know, if you think about this it's a pretty insane rate of growth rate within 3 months, you know, we're nearly at a hundred X growth in the project. And so, you know, what the first five Ford gross it was no problem. Like we're set up scale arbitrarily large scales, but you know in principle right? But in practice our server started grinding to a halt from all of the

communications are are disk drives were rapidly. Filling up with these yo, many different simulations that each each of which were a hundred gigabytes. And so, you know, in addition to trying to come up to speed on the science and be aware of what's going on. And where are the most high Valley area that we could contribute, you know, we were trying to keep things from falling to pieces under the load and so we were trying to the floor on the cloud and we were trying to shuffle date or around like mad to keep the discs from filling up on Arch on Primus servers

and you know our first tranche of data again even in its most strip down and compressed for ms200 Terry B. That's probably a factor of five smaller than the intermediate State we were in. So it was quite the crunch extremely grateful that we got linked up with you guys and you were up for her for stepping in and lending a hand. Yeah, I think you said you grew likes it initially, you had 30,000 users that were signed up and then it grew to like a hundred thousand years or so. Each one of those to bring a certain amount of processing power which is

fantastic because that means you can take all of that data and process it. Getting the data to be a hard drive, some things like that. Right? So this was a major challenge someone way of thinking of all the different structures that are protein, dumps, right? And so each simulation is effectively wandering around some part of this face structures. And you're sending us back to PS4 regular analogous to trying to build Run to the grocery store and what not. And then

I'm taking with the car now G. If we suddenly have a hundred times more explorers in our Fleet, all sending back data. We need a hundred times more parking spaces in your storage units for forever. And so petabytes in half, you were able to put it on our disposal was just a bunch of. I read that a hundred terabytes, the previously seemed really big, but suddenly seemed to quite small, given the an hour that was flowing into our system, the first couple of weeks when we have

ground like 400,000. Yeah, well. In Gregory's, I mean, there's a lot of things obviously, you know, for those of us who've seen a picture of this in the media, right? It's the, the gray ball with the little red fuzzy tips. That looks really cute. But you know, through our first meetings, you know, you talked about, I remember that the demogorgon right and he has the ability to Matheson this really interesting numbers to come out of it, things like being able to see something to the effect of a millisecond and even hear you when you mention point in time, but just be a

millisecond versus you being able to do a tenth of a second in a kind of explain what your ability to do those things. Really impacts the result that we can find. Yeah, so starting out with the technical side, you know, what happens with these simulations, as we have a three-dimensional representation of a protein and its surroundings and what we're doing over and over and over and over and over and over and over and over again, is asking given how each of the atoms in the system pushes and pulls on the others. Where are they going to be some small time in the future, right? In,

in order to avoid on physical Behavior, like atoms jumping through each other. We can't ask you, where has he gotten going to be a second in the future or asking? Where is he got him going to be 10 to the negative 15th parts of a second in the future. Right? And so if we have to want to build up to, you know, a second time scales, it's like, you know, a billion billion more than Billy billion-year time that we have to do this. So it's a lot of numbers, right? And you're so. So the upshot of this is that these calculations are extremely expensive. And so if you want to

simulate just a microsecond -6 millionth of a second then, you know that I can easily take hundreds of years on the desktop computer. Right? And so, you know, using clever computer science and physics approaches and bring many computers to bear on these problems making you Ceviche. And now we're at a point where we've made a major advance in the field where most paper is in our field will have about a microsecond dissimulation in a must-have papers about microsecond of simulation for a protein. And so what we were able

to do with a scale that we were on where you know, I've cheap. She now even very conservatively, we had five times the performance of the world's fastest supercomputer the time all working on this one set of protein from the virus that has far fewer components in the human body is now say, okay. Well we're going to get milliseconds the thousand times. More stimulation for each of the Systems that were looking at looking at it. So looking at most of the proteins from the virus on this time scale, you know, and I are getting our our first you know major chunk of data has 1

seconds of data thousand times more data than a typical good simulation paper in our field. And so as you can imagine this with gave us Insight that was completely inaccessible to anyone else without the access to this amazing resource. That are our Volunteers in Yeah, well, since you bring up papers Greg folding at home has it's written a few papers here recently. Can you talk a little bit about that so that I can go and just how much you guys have contributed to the project has always been extremely

productive different fields over the year. When you're one of the exciting things about your having so many people working together on this very, you know, pertinent set the closely interrelated problems with the viruses that we were able to move really fast. So you don't normally go to take a year or two just to come up to speed with a new problem. Get the lay of the land before even like taking on, you're the first real deep steps in this case, you know, and then about a year, we were able to go from. Wow. There's a flu

virus that we've never heard up the floor to our first. Major paper in nature chemistry, which is one of the Premier journals in our, our feel that this has been really exciting because they are in addition to describing the, the world's first exascale computer for the first computer that can do a billion billion operations per second is just insane. We were also able to get a lot of insight into how the virus works. It's the one of our main focus. Brian mentioned, is what we've come to call. The other covid-19. Demogorgon

the technical name for. This is the spot like that. When you see these images of the virus, you know, you got this ball with all these spikes coming out of it. Each of those spikes is a complex of three, identical protein range with a three-fold symmetry of quite a bit, right? Because this is the protein that is responsible for attached. The proteins on human cells and initiating infection, right? And soda perform that function as you can see it's out on the

huge liability for the virus because it's one of the things that are immune systems have a great chance of recognizing when the virus arrives in our bodies and you're triggering a strong immune response that clears the virus before it even has a chance to get into any of ourselves and really take off growing. Right? And so, the virus has evolved a number of defense mechanisms in the one that we were most inspired by if you will. Is that your the structure of the spite that you see in these images actually closed up on itself, right?

Kind of like a turtle, having its head and it shall write. And so this is a great protective stance for the virus because it protects the most sensitive bits that the need to engage with a protein bunch of themselves from being. Of course, being buried that means that they can't actually, you know, get at their prey. They can't eat with the fertile, right? They can't find that the semen protein called a stew. And so in order to do that, this thing has to open up my. And so, with the threefold Symmetry, we kind of imagined. This being like the mouth, one of these demogorgon monsters

from the television series, stranger things. And we really want to know what that looks like. Great because mostly what anyone has managed to see experimental, he is this very common closed up structure. And so using our our computational power. We were really able to have to see a lot of really traumatic opening up of this Spike which was fascinating because of all, you know, is that it also looks like this, you know, what's the surface that you have available to Target with therapeutic, but if you know that what it looks like, once it's opening up, you suddenly have all these

previously unknown sites to Target that we call kryptik sites. And fascinatingly, you know, after we started sharing about this new on social media and stuff, and showing people data so that they could start taking those ideas as quickly as possible. Even before publication. Some people did start reporting that. They have discovered antibodies, that neutralize, the virus, and surprisingly behind in these deeply buried regions that no one thought of that, I could get to answer. This is a nice corroboration, that has an example of one of our predictions

and we've done your similar things with many other protein from the virus. I mean, it's it's also cool. I mean like learning to do, you know the name of NS p16 and wanting to read those things? The one that really what? I was reading all the things that came through and one of them got a lot of attention was around this idea of face separating with RNA. Like I just this year I learned what are Nae was. We've always sort of notification but I learned everybody else. But Arnie was with the idea. What is what is that paper about the one about face separating with are in a sort of

explain that to me? So it's been common to think about the proteins, kind of like Lego blocks, right? And they go off and they attached other proteins, and they do different things. And you know, so I sell, is this bag, you know, that's full of proteins and some of them structural roll. Some of them are going around different parts of the cell, you know, catalyzing different chemical. Has been emerging in the last two years is the things are, even, even more complicated than,

than that, right? And that some of these different types of molecules. Can I kind of like oil and water, right? Or instead of being all mixed up and going about, you know, randomly for lack of a better word and different different areas as well, mix system like oil and water. They actually separate and that this is actually really important than that biology has likely use the strategy. Tune able, you know, compartmentalization of different functions. And so this brings up the 10th wising idea that okay. Well we have these, you know, viruses that are

trying to invade our cells and replicate themselves. And so, there are making all kinds of viral pieces inside of the salad. If you think about them as randomly distributed throughout the cell, it's really perplexing. Like, okay. Well, how does that All these random pieces assemble into working viruses. Right. Kind of like saying, okay. I'm going to dump a bunch of car parts in my garage. And then I'm going to shake it up. And so you think everything is random but despite your belief that things might be random is suddenly going to bring together into half a dozen working cars, right?

That doesn't sound appetizing possibility that the ability to face separate than have, you know, things group together spontaneously is an important step in that process. That sounds cool. Sounds like we need more axle flops of computing. To figure that one out further. Would you have a timely as super great timely? Question, from the audience, which is, you know, how do you use social media to share information and in certain ways gain trust? And I think this specific article that was on it was published by Nature. Cam is a great example of that,

right? Yeah. I mean this in a really cool thing and part of why we were able to have so so much impact as that you have folding at home has been around for a long time. There's a lot of people that I've participated this whole time, you know, where we're pretty active on Twitter and Facebook and our our blog. So we have good visibility. And yeah, there's a lot of people that have been part of folding at home, you know, maybe they were exposed to it in a high school science class where everyone is folded on a laptop computer or whatever, right? Or are they got involved at some

other point and then maybe maybe Drift Away. What is really cool is that you know in in response to the pandemic we near like okay, like you normally one might be tempted to be a little defensive with scientific work at the bleeding edge. Right? Like you want to make sure you do at the name publish it so you can get credit but especially you think we've been trying to promote open science science go as fast as possible and we'll work the credit and all that out at the end. And so we we really ramped up with the pandemic and where you're sharing information via Twitter and stuff

you do through throughout hopes. That people would be able to put it to good use and you know sharing all the state online that's been been really great. I think people really appreciated getting some, some insight into what we were learning about the virus and your what steps we were taking towards, enabling the development of Therapeutics and seeing that all the computer is being put to good use Yes, it's so great to add on to that. You know, we talked about sharing information. We talked about, just the massive amount of data that you guys are collecting. Can you

explain in a little bit of detail, the role that flashlight is playing in, in all of that work and what you're using it for our folding at home number of roles and you don't have to apologize to you. I probably won't even succeed in the sting them. All right, but yeah, one of the first things, you know, that's right up front is like I said, like we just had this date that we didn't know how to handle the right. It's just that the magnitude of what you guys gave us was literally a life-saver, right? Otherwise we were going to have to, you know,

start deleting stuff or are you moving it into places where be hard to retrieve? So, so, that was the first phase. But, you know, it's not it's not just a big hard drive. If you guys know it's, you do got a lot of extra. Abilities. In terms of compressing things in the navel. I need you to work on things, you know, it in place after like insane speed, relative to what we do. We have to have in place before with traditional spinning disks now is for us to take some of our work when the algorithms for folding at home to the next level.

What would happen in the past? We have any servers that you are basically Big, Ray's that you're bringing data from our volunteers and then store it until we're ready to operate on it ready to analyze it scientifically the typical work for low would be that you know, someone on the scientific team with set up a thousand simulation, right? They would add them to the queue on one of these work servers and they would kind of like check on whatever couple weeks for 3 12 months, right. And at some point where it's okay, you know, that's probably enough data for the scientific questions were we're

trying to ask now. Let's copy that day. Grover from the server to our computer cluster where we have no bunch of nose with lots of CPU, cores on gpus and enough memory to, you know, go and dig into the state and really analyze it. But, but in the meantime, what we've developed or these algorithms that we call, adaptive sampling were instead of just seen your letting me simulations go for a long time and then getting all the date at the end of that work, in a more active role, where we start the simulations. We let them run for a while and then we stopped them and analyzed all the

data that we've generated so far. And this is this map building exercise that I keep referring to where we're going to build a map of what we've seen so far and then we're going to use that map to side. Well, we're on this map. Would I benefit most from getting more data either to improve parts of the map that I think are important or to expand the boundaries of what the map covers for example, and now let's go. Yo, start the simulations from there and get our next batch of data. And when we've been doing this for a couple years now and our computer cluster, but for a variety of

technical reasons, this is really hard to stay a lot for folding at home. We need a lot more computer power to analyze all of the state of that would come in for more. Volunteers on the Fly and saw the flash player has really been enabling for that. For that always been real, architecting a bunch of pieces of building a home. So that we can you have. Our work server is immediately bringing data into the flash blade, which is also connected to our heavy compute machines. So that we can analyze it there in place. Make those decisions until the work server know which of the structures we found so

far from the last batch of simulations with patches of stimulation. Start that next batch of simulations guns, that we can make the most effective use of finite Computing resources. And you know it you actually mentioned part of this in your conversation where to get back to the real quick Greg, but we got to write question for the audience around, you know, people who want to get involved with a just that the business of this type of science. But also just like what is it to get involved with a? I and so Stacy, I'd love for you to go to share with people. What do you think of the skills of

the future as it relates to a, i Yeah, so I would say it's a it's a combination of quite simply getting enough data, right? Having access to enough data and then just experimenting using different programming language and different applications. Most of which are open source that you can begin to stitch together applications and programming languages to bring in raw data. Manipulated in the way that you think might be beneficial to what you're in result is going to be and then come out and visualize that data. So I would say in terms of

skill sets. You know, there's a lot of different. Horses out there online. And and quite frankly. There's a lot of different models that are already available in the industry that people can experiment with. So I would say in a roll up your sleeves, get out there. Start experimenting with the data. I'm going to make a plug for cagle.com and I'm going to let Greg talk specifically about cuz they do make their data in a break. They make it home. It's their data accessible to anybody wants to experiment with it and he can talk about where you can find

that, but there's also a lot of great calm and different projects that are available. I think it's just as simple as getting out there. Getting access to data experimenting with the data in a multiple different ways. And I'm getting really comfortable with models. Like I said, a lot of models have already been developed. So it's not like you have to build it from scratch. It just experimenting with those models and getting a sense of what they're capable of and what they're not capable of us and then pairing it with your data. And are you getting the outcomes that you fixed? That from

us, more industry broad perspective. That's what I would say. The skill-sets are more around programming, programming languages to manipulate the data and combining different data sets together is if my personal opinion and then I would, I would say last but not least, then we talked about the massive amount of data about it getting data to the right place at the right time. I definitely think there's a place where the cloud, you know, a big component, a folding at home is actually even hybrid or some date is going to land in the cloud and then some date is going to come back on from

making smart architectural choices that give you access to your data, portability of data, is going to become more critical to organizations and not being walked into one. Or the other is going to be really, really critical. So we see a lot of demand for what's really enabling that portability being around Kafka and kubernetes containerization. For example, We love these questions from the audience to keep them coming, you know, from my perspective. I love the ability to be able to architect, you know, extremely large

systems performing Enterprise AI in. And, you know, do you know actual biotechnology, science Ai? And all the things are between like a is totally different use cases and, you know, machine learning, whatever you want to call it today, but I think it's really cool about Greg's job and the other, you know, the flip side of this and we can talk to him a lot. These doctors who have these backgrounds and very unique science has to also bring up things that are heavily mathematical dimensions of doctors, sampling algorithms. You know, that that's not necessarily his it wasn't part of his

thesis or dissertation, right? But it's really maybe it was really cool stops being combined, those two techniques to get these types of outcomes of Greg. You want to talk about your perspective here. Yeah, I think it's a really exciting time, you know, and briefly in our field, kind of at the interface between machine learning and particularly deep learning, and physics, because we have a lot of lot of data coming in and your great desire to make sense of what's going on. So I was just one example, you know, we've been getting into deep

learning to hear take all of these different data sets. So, for example, we've got simulations of the spike complex. I talked about from the Stars covid, 2 virus, but also other coronavirus MERS and nl63. The original Stars. One buyer, impactful virus compared to these other viruses that while scary in their own, right. And you did not become as much of a global concern by any means as the Stars. I'll get you are devised in another recent recent nature. Communications paper is architecture called dip Nets where we can take

in structures from the different simulation and through through deep learning. We can a coarse grain that data like try to come up with a simpler representation of that stuff. I had a long story shortened, the process were able to force it to say. Okay. Well if you're going to do that at work, you know, you need to figure out what the most important differences between these different data sets are and that's been an extremely powerful way of learning like, okay. Well, what is it that unique about stars Covey, to that distinguishes, it from these other coronavirus has and

help enable it to be calm This Global pandemic, that none of us are going to forget. And I'm already, you know, preparing for my my grandkids or whatever to, you know, ask me about what was life, like during the 2020. You know, for their, their essays in the future, right? And so this has been been just usually enabling and it's so easy to get into with things like pytorch withdrawal publicly available and datasets like ours that you can find on a service call to OSF in through the AWS public data sets program.

I'm going to tell him I was hanging out with you right now. So your partnership, you know, we know that you're working with tons of great companies out there that are providing your gpus and CPUs and networking and we're privileged to help help. How are you with yo really fast simple, you do unstructured storage for that same type of use case. It's been fun to work with you and you know, I go ahead Greg saari. And, you know, we we want to challenge everybody and I still run

folding at home on my computer over here. I can, I literally can pop a breaker, almost once a week. I'm looking forward to the winner. Like when I could use it to help heat the home, you know, it's a, there's they still have hundreds of thousands of users. They're not just solving covid-19 resolving. All these other things. I mentioned Alzheimer's, heavy metals, buildup in the brain. And I'm like, does that have to do with the protein? Is the protein fault, you know, so it was like, you know, it's like it's so exciting to have learned all these things about science that I did not pay

attention to years ago. So I thank you for helping me. You'll get get past my poor, my poor education and settle again, everybody jump in, we're building science, careers were solving problems. Greg is doing amazing things. The world's largest distributed supercomputer. You're going to see it again what they're going to set a record of get right so really cool stuff everybody. Thank you so much for joining us again, Stacy. So fine, so fine. Thank you. The more people, we can get involved, the merrier even with all the computers we have, we're still compute limited the problems without

hard and the result is that important. So we look forward to continuing to work with you and everyone else wants to get involved.