This episode features an interview with Andrew Fong, Vice President of Infrastructure at Dropbox. Previously, Andrew has served as a senior system administrator at a number of technology companies including YouTube and AOL. On this episode, Andrew talks about how to use data storage to create better workflows, the future of cloud data storage, and much more.
This episode features an interview with Andrew Fong, Vice President of Infrastructure at Dropbox. Previously, Andrew has served as a senior system administrator at a number of technology companies including YouTube and AOL.
In this episode, Andrew talks about how to use data storage to create better workflows, the future of cloud data storage, and much more.
--------
How you approach data will define what’s possible for your organization. Data engineers, data scientists, application developers, and a host of other data professionals who depend on the Snowflake Data Cloud continue to thrive thanks to a decade of technology breakthroughs. But that journey is only the beginning.
Attend Snowflake Summit 2023 in Las Vegas June 26-29 to learn how to access, build, and monetize data, tools, models, and applications in ways that were previously unimaginable. Enable seamless alignment and collaboration across these crucial functions in the Data Cloud to transform nearly every aspect of your organization.
Learn more and register at www.snowflake.com/summit
Steve Hamm: [00:00:00] so I wanted to ask you a little bit about yourself.
To start. I noticed that you've been at Dropbox for about eight years, which is pretty long for a technology startup or Silicon Valley company, so you must like it there. Why have you stayed so long?
Andrew Fong: [00:00:16] That is right. I had joined Dropbox, , and in 2012 at the time, I think the company was about 120, 130 people.
I moved over from Google and I have found. Dropbox to be this really unique experience of this intersection between technology and people, and that has created this really unique culture for me. That's actually the main driving reason that I stayed. I would sum it up as when you work in infrastructure like I do, it's very hard to find.
A scale and a reach in, in most companies that you're afforded to build on top of. And Dropbox had that. And then we also had this incredibly talented technical set of, of engineers. And that intersection for me really just jelled. And I've just had a fantastic time bouncing around between technology projects along with and working with a set of exceedingly talented engineers.
And that's the reason I've stayed.
Steve Hamm: [00:01:09] Yeah. You mentioned the culture. What's so unique about it.
Andrew Fong: [00:01:13] I find that we really try to put the technical efficacy of projects in a level that. Really resonates with me. Um, we try to make sure that we make the right technical decision decisions coupled with the right business decisions.
We make sure that we have the right discussions and we have them upfront and we're pretty Frank about them. , and we really try to remove the, you know, all the inherent biases. Around why you should do something or why you shouldn't do something and look at it from the lens of what's best for Dropbox.
And so that we used to have this value of we not I, that value really comes through for me. I find that cultural value of like, of just trying to do the right thing overall for the company to be really inspiring.
Steve Hamm: [00:01:53] Yeah. Cool. Cool. Cool. So you know,
the name of this podcast is the rise of the data cloud. And that's obviously our big theme. And yet, you know, I note that in 2015, Dropbox migrated. From the public cloud to running it. So data centers, kind of the opposite of the trend that we are focused on. And so why did the company do this w and what were the results?
Andrew Fong: [00:02:20] You're correct. We did migrate our data, our storage platform layer away from the public cloud. I will. I do want to take a step back and level set with how Dropbox is architected. Dropbox has two main pieces of content. One is the metadata about files would, this is the access, is the ankles, the creation time, who owns the file, et cetera.
The file name, where's it located in the directories? And then also storage content. And from the very beginning, we've always been in a hybrid mode. We've always had our own data centers or on footprint as well as leverage public cloud. So we've had that competency and we wanted to build on that competency and we saw ourselves as a company that was going to be fundamentally powered by data.
This is an, at the time, I think we had roughly 600 700 petabytes of user data in the public cloud, and we said. We need to have control of this data. We need to be able to make and find efficiencies in this data. We need to, to analyze this data in any way. We want to analyze this data. We want, we really saw data as a core competency that we needed to have as a, as a business.
And so we took, we started that journey or about migrating out of the public cloud in 2015 for storage. We still do have a footprint in the public cloud for other, for other things. , but if this is primarily around storage, and we just. I felt it was a really key part of the business and at the scale we were operating it, operating at that we could find those efficiencies.
I got to answer the first part of your question. Yeah,
Steve Hamm: [00:03:46] yeah. Well, let me, let me pin that down a little bit so I wouldn't understand. So you moved, so all the storage, your, your, your customer's data, you moved into your own data center.
Andrew Fong: [00:03:57] Correct. We moved about 600 petabytes, 700 petabyte at the time.
Steve Hamm: [00:04:01] So what's the advantage of moving that into your data center?
What efficiencies or ability to analyze the data to get out of that?
Andrew Fong: [00:04:11] When you think about building a storage system, you have to think about the customers and the use cases that go on to that storage system. And when we designed our storage system called magic pocket, I may refer to it as MP, but it's the full name is magic pocket magic pocket is designed.
As a purpose built file system and storage system, block storage system, not file system, block of storage system for Dropbox. And what this means is we don't actually support all the APIs that the public cloud was supporting from their block storage system. We built a storage system, which was purpose built for the workload that we had.
And when you do that, you can build an end to end vertically integrated stack. You probably don't want to do that if you have one terabyte of data because you're not going to get any economies of scale. But when you start to look at, you know, data sets approaching an exabyte, and we're well over an exabyte of user data at this point, when, when you have data set sizes that large, you actually can build a vertically integrated stack that is much more efficient than a general purpose stack.
So think about this as. No. A lot of times in the ML and AI world, you hear people saying, I'm gonna use a GPU because it's more efficient. I'm going to get a better, I'm going to, it's going to be faster. It's going to give me better performance, can be cheaper for, for a number of cycles I need. This is a similar sort of play where we're actually able to do that with storage.
Steve Hamm: [00:05:23] ,
yeah. Okay. So what parts did you leave at the club would put your business with, with data?
Andrew Fong: [00:05:29] What date did we live in the cloud? Things that we've left in the cloud. Have been international block storage actually, because as soon as we were talking about you have to have this economies of scale and you do have users and customers that want to have data stored outside of the, the, , outside of the North Northern America, which is where.
Locations are storing data. And so we leverage public cloud in Europe, in Tokyo, and in Australia as a way to extend our footprint and allow local data to be, to be stored within country, within region. And that's, we look at that as a feature that we can provide. We built it as a storage system that has an abstraction that allows us to route data blocks anywhere we want.
So there's not a, we're not beholden actually to have any single one storage system store these blocks. We can actually store them. In our magic pocket system, we can store them in any number of public clouds if we decided,
Steve Hamm: [00:06:21] so let's get to, let's get to snowflakes technology because, um, I know you use it.
So if you could explain how do you use it, that'd be great.
Andrew Fong: [00:06:31] so if we're thinking about how do we use snowflake and how do we adopt it?
What we look at is we've had a very robust analytics data pipeline. But at the end of the day, you have to service the business user as well, not just the engineering footprint user. And what we found snowflake to be very good for as a use case for us, is actually to slice the data and I'll give the give interfaces to.
A nontechnical set of users. And when I say nontechnical set of users, I want a level set that I'm talking about in the engineering team at Dropbox, which is going to be much more accustomed to going in and writing a bunch of code to analyze data and be comfortable with that. And so this is going to be a bridge into.
The, into the general business operations units into the teams that needed to do financial analysis. We need to be able to give them interfaces that allow them to make performance queries that be able to visualize that data very quickly. , and generally give them an experience that's not, uh, that's not what an engineer would have.
When you give out an experience, it's much more from, I would say, consumer, but like a true end user. We need to look at it from that perspective and we really found that stuff like gave us the ability to do that from, from that perspective.
Steve Hamm: [00:07:39] Yeah. Let me see if I understand you. You moved your, your customer's data.
Into your own data centers, but you're using snowflake for business analytics. Isn't that a way to put it?
Andrew Fong: [00:07:52] Yes. So we use snowflake for business first is first set of business analytics, , and we still have a data Lake and a data pipeline that's in our production systems, which runs both on prem and public cloud.
And that all funnels into a centralized public cloud sort of storage system, which then routes some portion of that data into the snowflake systems. Okay.
Steve Hamm: [00:08:12] Cool, cool. Now you mentioned engineering, general business and financial analysis. Can you kind of walk through some of those? How exactly are those functions using the snowflake, like technology and using analytics based on that technology?
Andrew Fong: [00:08:28] The primary use cases for this are going to be routed around understanding, where users are in the world, that type of data.
It's not going to be cutting sort of the ML and AI side of the, of the analytics, which is going to be much more powered on the production side.
Steve Hamm: [00:08:46] so when, when we think about cloud data, we think about some of the sharing capabilities about specifically the ability to break through the silos. Organizations often have within, you know, within their own structure. Is that something that's useful to you about the cloud data warehouse in the cloud data platform?
Andrew Fong: [00:09:08] , we definitely have a strategy internally where we want to democratize, democratize the data as much as possible so that we can make high quality business decisions based on data. So I think about this as. No, it's a tiered architecture. You must have some repository of all the data. You must govern that data in some way so that some of these metrics are actually going to be certified metrics that we can actually push through pipelines all the way up to the CFO financial dashboards or all the way over to the product teams that need to make very clear cut decisions on whether they should launch a product or not launch a product.
And so as you move up or down that hierarchy, you're going to have different types of interfaces that you're going to need to have for different types of users. , and so when I think of that, how will you leverage snowflake? It's going to be much more on the top end of that stack, more on the CFO side.
And then as you move down to the product product analytics and keep going down the stack, you're going to probably shift more into, I would
Steve Hamm: [00:10:01] say.
Andrew Fong: [00:10:01] A user base that wants to consume a tremendous amount of data that's very unstructured. , this could be crash reporting logs, for example, like our analytics pipelines stored that type of data as well.
It's not just purely a set of business making decisions. It could be we need to use this data to, , to analyze crash rates across all of our desktop clients. Which number in the hundreds of millions to billions.
Steve Hamm: [00:10:24] Yeah, and that's unstructured data, right? I mean, you're just putting that in the data Lake.
Andrew Fong: [00:10:28] That's very unstructured data.
Steve Hamm: [00:10:29] Yeah. Okay. Interesting. So, um, you know, I have a question about your business. I mean, I, I'm a great admirer of Dropbox. You know, I also wonder about it because. You know, cloud storage provides tremendous utility for businesses and for individuals that need, but put on the face of things.
It's a commodity business. So how does Dropbox add value and achieve high profit margins? You know, keep it stock up and all that kind of stuff. So
Andrew Fong: [00:11:00] from a, if I look at the types of data that we store from a customer perspective, it's just an astronomical amount of data. It's one of the largest or paraphrase of data in the world, most likely from a, from a, from a business and consumer perspective and a unified way.
And as customers put more and more data into it, they think build workflows on top of that they can, and we can enhance the value of that. I would, and the simple example I use, my wife is a lawyer. She sends files back and forth all the time for contracts, and she read lines them. And then she's always constantly trying to figure out what's the diff of contract a versus contract B?
Did I send it to the right person? What if that person's out of the office? How do I get a response from the lawyer at the other company? , and so when I think about. Using Dropbox, right? This provides a workflow that's that we, that can be built on top of that data. So she can look at the diff, she can get the revision history, she can understand who's looked at it, who's modified it.
She can look at the commenting history on there and it's document type agnostic in some sense. It's not just a word document. You can do this with a PDF. She can look at it on her mobile device. She can look at it on her, on her iPad. She can do it as she's, you know, on union San Francisco. And it gives her a way to free herself from the sort of desktop application that was there in the, in the previous world.
And so these smart workspaces that we're looking to enable, give us a bridge from what was file sync and share in a very simplistic manner, which you're talking about the commodity side, much more into a bridge around collaboration where there's another party and it's not just you using your document.
But if someone on the other side as well. So I use like to use that example because I, it hits home for me when I watch her and her workflows. , and she does, you know, she works at a place that doesn't use Dropbox. I constantly tell her, you know, if you, if you had this, your workflow would be much, much better and you'd be able to collaborate with a much wider set of people in a much faster way.
So is that efficiency she would get from her workflow as
Steve Hamm: [00:12:49] well? Yeah. So in this case, is Dropbox a platform but these other applications are built on top of, or do you actually build. You know, vertical or horizontal apps into your plan, you know, into your technology.
Andrew Fong: [00:13:04] So we have integrations now at this point, within the desktop application surface.
So we have partnerships with zoom, for example, where starting a meeting around a document, it's something that you can, that you can do straight from the Dropbox client. You can build other integrations, , with some other partners, I believe at last name is a partner where you can tie in JIRA tickets, you can tie in confluence right back to the content, right?
And so now you have the single place where you in single pane where you can go and you can create workflows as needed. Whether, and so you're not having a thousand browser tabs open. I know that's something that I constantly struggle with is I have a million websites open for a million of replaces, but I still can't find that one piece of information I need.
And so from our perspective, if it all starts with content and we have this repository of content and people are working on that and collaborating on it, that should be the focal point and the thing that everything else rotates around.
Steve Hamm: [00:13:52] So you've talked about these new capabilities that you've built on top of, of your storage capabilities, these, these truly applications and workflows. And I'm just wondering, have you used data analytics either to help you kind of design or improve those new capabilities.
Andrew Fong: [00:14:15] I would go as far as to say that every single feature we develop, every single experiment we do is all about using the analytics pipelines.
We want to make thoughtful decisions about what we launch, what we don't launch. And then we also want to use them. You know, if you take it to the logical extreme rate and the ML and AI side of the world, we want to use that firm for surfacing there better data to our users as well about what they actually could take advantage of.
Um, so in our tray, in the menu bar populating. , recent documents around a meeting, for example. , you can use the debt, simple heuristics where you can build some, some more sophisticated MLN AI models around that. So that's like on one Luxtreme stream of how we're using the data. And then in the middle there's the generalize.
We've launched a product, we understand sort of the AB testing about how is it doing, what was, what were we seeing more active users being driven. , from solution a versus solution B. And so we ha, we take a very rigorous process , from that perspective, very standard sort of industry best practices around. The AB testing and then making sure we're using analytics and the proper way to actually build models around understanding what the users are actually doing and what behaviors they have when they, when they enter the site or they're using our Dropbox clients.
Steve Hamm: [00:15:22] Yeah. So do you consider your, I mean, the technology that you run in new and data centers, you, that's basically, it's cloud based as well as the, it's not how you define it, but it's private cloud.
Andrew Fong: [00:15:33] I would define what we run as. I would say that we are a cloud service provider.
We're not providing utility computing. We're providing a cloud service, and by that definition, I would say that we have built a cloud. Our cloud is around the Dropbox ecosystem
Steve Hamm: [00:15:50] though. How would you differentiate between the cloud and the utility though? I'm not quite getting that.
Andrew Fong: [00:15:57] The way I would differentiate it is when I think about this major CSPs that are providing raw compute or raw storage with no value out on top of that, that to me is that utility.
Think of it as electric company. You get electricity coming in, you can buy a computer, you can buy some CPU, use that CPU. That's the way I would define the AWS cloud or the GCP cloud. And then I would say that that Dropbox is a cloud service provider in the sense that we provide a, we have to build a platform and then we build a platform and we build this platform called Dropbox on top of that.
And that is a service that we're out outing back out as a cloud. As a cloud is a, is a public cloud provider. In that sense,
Steve Hamm: [00:16:34] you have your own cloud. Yes. Okay. what are the biggest challenges you face on your job and how does data help you solve them? The
Andrew Fong: [00:16:44] biggest challenges that I face within infrastructure.
A lot of these are going to be around launching new technologies and at the scale we operate. Data plays a very big role in that. For example, we have millions of hard drives in production. These hard drives generate a tremendous amount of data about the hard drive themselves. We need to analyze that data.
We need to understand the failure rates of hard drives at a massive scale. We need to understand and be able to predict when hard drives will fail before they fail. So it allows us to take proactive action against them. We need to understand how data placement and locality around the world plays back into our performance metrics.
We need to understand how, what the applications are doing in production from a monitoring perspective. So we stored incredible amount of monitoring data as well. Just about sort of the performance characteristics, the latency characteristics, how was the application performing once it's deployed? So data is.
Fundamentally ingrained in how infrastructure works, because we have to use this for every piece of decision. We may wait from financial models all the way down to, , is, did we, what was the performance of that drive when we stored this bit of data on it?
Steve Hamm: [00:17:57] So, yeah, back to the drives, had the performance of the system. You collecting a tremendous amount of data. You're streaming it, you're ma, you're essentially monitoring these working machines. So it seems like this is one of those great examples where it's really a programmatic.
People aren't involved in making decisions. You have a system that is tuning itself in real time using data analytics. Is that the correct.
Andrew Fong: [00:18:27] At a high level. Yes. We have a stream of data that comes out from the drives that's put into a logging pipeline. The logging pipeline puts that into. Into a effectively and actually in the same data Lake that everything else goes into.
And then there are some jobs that run across that to look at the types of, I look at the types of failures and then generate tickets as needed back to the data center teams replaced drives. And then you can also do things in just how we work with our vendors from the drive side is like we want to look at those drives and look at the failure rates over time.
Look at if they're matching up, what the benchmarks that we had when we qualified the drive looks like. So we're using analytics from that perspective just to always make sure that this is the case. Like what we bought is actually working the way we expect it to work and that we're not missing anything about the performance characteristics of the system.
And so some of this comes from the analytics pipelines and some of this comes from more of a time series database type pipeline.
Steve Hamm: [00:19:22] Yeah. Yeah. I wonder to see, one more question I want, I want to ask you to step back and be a visionary for a moment
Andrew Fong: [00:19:29] here.
Steve Hamm: [00:19:30] So looking ahead over the next decade, how do you think data and data analytics will kind of change the world of business and even the world for people?
Andrew Fong: [00:19:42] I think that right now in the world we live in, I, you know, we're in the midst of a. Unprecedented times in this, in this global pandemic. I think that there is a huge amount of data and value as people start to, that will be used as people start to work from home more and more and more. And we are even at Dropbox looking at that and trying to understand the user's behavior because the user's behaviors, and I know my personal behaviors, I'll speak only for me, have changed substantially.
Just since I've been working from home over the last few weeks or the last eight weeks, and I fundamentally believe that we will see evolutions in products. We'll see products that no one thought needed to exist, and that will all be driven from the understanding of the behaviors we see via the analytic, via the business intelligence, the analytics we're collecting.
We'll use that to make new and better product decisions around just even in Dropbox, like how does Dropbox work better and better suited for the. For a work from home workforce. , I think those sorts of things, those sort of problem statements are going to pop up more and more and more. I think that's probably the, that clearest advantage from a, from, if I think about Dropbox from how data will change over the next five to 10 years for us.
For sure.
Steve Hamm: [00:20:49] Yeah. Yeah. Yeah. When I think about it, you know, if people don't return to normal, if offices kind of become obsolete. And they might be in some, in some cases, you certainly don't need the old storage lockers, you know, the old file cabinets and things like that. And it could, it could kind of drive the last wave of digitization, you know, for going fully digital.
And then, you know, flying the coop out of the office so that, that could be a pretty interesting transformation. And. Cloud storage is an essential element of it to make that work. So it's pretty cool. Andrew is, is there something that could take us a
Andrew Fong: [00:21:32] little deeper
Steve Hamm: [00:21:33] on the cloud data platform?
Andrew Fong: [00:21:36] I think that the biggest piece, I mean, the, what snowflake has given us has been really getting the fundamental, biggest use cases have been really around financial reporting data, I think has been the biggest win for us in the short that I, that I'm aware of. . I've gotten a little bit more involved in the last couple weeks, but that's, and that's been the highlight over the last couple weeks has been just how much we've used it for financial reporting and making sure that we can surface the right information there.
Yeah,
Steve Hamm: [00:22:04] yeah. Maybe give a little more color on that. is there something about that technology and the way it's delivered that enables them to do things that they couldn't do before or couldn't do as well before?
Andrew Fong: [00:22:16] When I think about the financial reporting data and the ways we're building dashboards using snowflake. The really big value add for us has been how simple it has been for the business intelligence teams to be use it to create these dashboards. Our current systems and our other data lakes and the other systems we have powered or powered by a little bit more, a little bit more complicated systems that are gonna require deeper knowledge of the data sets as well as technology stacks.
And what this has allowed us to do is remove the barrier to entry into data, as well as provide a way and a way to create a gate to allow us to filter the right sets of data into it as opposed to make a user base have to understand. Yeah. Hundreds of millions if not billions of metrics. And this really gives us a way of curating the metrics pool and creating a very simple, simple interface, which I think you know, is, is a company value.
Ours is around simplicity, which is like a huge value add for the, for the, for the user of the system. Complex systems almost never result in someone producing the result you're looking for. So I think that stuff like is really given and really nailed it for us on making sure that our user base is able to get.
Exactly what they want by not building something that's over complicated for them, which is in some sense the way our rest of our data lakes work.
Steve Hamm: [00:23:37] So in a sense, it's democratizing data.
Andrew Fong: [00:23:40] It really does help democratizing data democratize data for us. Yeah.
Steve Hamm: [00:23:44] And it doesn't require a data scientist to do sophisticated work.
I guess
Andrew Fong: [00:23:50] I would say I would go in one further and say it does not require a deep engineering expertise of the product to make it work.
Steve Hamm: [00:23:55] Oh, okay. Very good.
So here we are in the middle of this crisis that's impacting people's lives in so many ways, impacting businesses, the economy, all of society. How are you seeing that impacted in, in Dropbox
Andrew Fong: [00:24:11] at Dropbox? Covert. It's been, it's all encompassing as it is for almost any company in the world or all companies in the world at this point, I think about as theirs.
Two or three major work streams. One is around people just making sure that people are safe, making sure that people have the tools and technologies to be able to work from home. We had an excellent con strategy from our, from our, from our its teams, technology teams on the internal side. Around being cloud native and being cloud first as a strategy, and so that allowed us to make a move to work from home incredibly seamlessly.
The entire workforce basically was home the next day, and when we declared a work from home and everything just worked. Now my role, I'm on the technology side is head of infrastructure. This journey for me actually started late January. our head of supply chain came to me and said 2020 is not going to be a normal year.
And for those, I don't know, supply chain typically has pretty far reach out into Asia. We actually do a pretty deep integration of our stack all the way down into the vendors on the OEM side. So pretty strong relationships with the Asian companies that are building physical machines. So we started looking at supply chain and we started looking and figuring out, okay.
How do we get to tier two, tier three commodities? Because we realize that things such as sheet metal even are going to be a scarcity around manufacturing. And so we've been diving in at that level from the infrastructure side. So this has had a fairly large disruption, a fairly large, um, amount of, uh, work put in from the supply chain team just to, just to keep capacity coming.
And I think every company in the world that with the supply chain right now is, is being impacted by that. , and on the third, I think is, you know, looking forward, right? Like we have to have a work stream around what does work from home look like and how do you keep, so the mental health aspects of, of the workforce, how do you make sure that people feel cared for, feel connected, are able to maintain engagement back to the organization.
Um, and so to me, those are the big three things. Those are the things that we're really thinking a lot about. right now.
Steve Hamm: [00:26:08] back to the supply chain thing. So you basically, you have a bunch of data centers, you are buying all of the storage equipment, the computers that networking devices, all that kind of stuff, and outfitting them.
Do you reach all the way back into the component vendors who, who put that together? It seemed like you were suggesting that, but that's really, I never would have
Andrew Fong: [00:26:29] guessed. So we reach, um, we maintain relationships with about seven different, seven, eight different commodity vendors. , for key components.
Think CPU. Uh, we were the first company in the world to deploy AMD at scale. For example. Um, we were the first company in the world to deploy SMR, which is new drive technology at scale in the world. so we have very, very deep. Supply chain and vendor relationships with the major drive suppliers, the major ship suppliers, uh, major Ram suppliers.
, we have that. And that's part of the portfolio. And we touched on this earlier about sort of the vertically integrated stack. In order to get the economies of scale, you have to have that level of integration all the way up and down the stack. And so we maintain those relationships with commodity, with the commodity teams.
In this case, in sort of covert 19, we had to actually reach a little deeper into the supply chain so we could understand exactly what the impact of delivery schedules was going to be, exactly what the types of things, exactly when and where things were going to show up. , every, once every. Every week there's been something new shelter in place in the Bay area, put pressure on integration facilities here, which we weren't anticipating because most of the supply is coming from Asia.
But you know, we do some final assembly in, in, uh, in the Bay area for some of the data centers here. That caused a disruption in supply chain for a day or two while we had to figure out how are, how do we compensate for that from an, from an essential workers perspective. , so there is a, the supply chain aspect is a, has been a very large focus for the last two or three months.
, just to make sure that we're able to, uh, maintain the runway, maintain all of the things that make Dropbox stay powered and run.
Steve Hamm: [00:28:04] so Andrew, I just want to thank you for being with us tonight. I thought there was some really cool insights. A lot of stuff I just didn't know. So, uh, thanks for sharing it.
Andrew Fong: [00:28:14] Steve, thanks for having me on. I really enjoyed talking about data with you today and until next time, thanks.