00:00welcome to the a 16z podcast I'm Michael
00:03Copeland the place where Apache spark
00:05was born UC berkeley's am pleb has not
00:09just created a major open source
00:10software platform it's spun out more
00:13than its share of groundbreaking
00:15companies full disclosure a 16z is
00:18invested in three of them so how did
00:21they get there how is open source and
00:23the amp lab approach reduced the
00:25friction between student and faculty
00:27ideas and launching them into the real
00:29world co-founder and director of the amp
00:33Michael Franklin joins a 16 Z's Peter
00:36Levine to discuss the amp lab model and
00:39their own relationship as an academic
00:41and an investor Hao Yan Li also joins
00:45the discussion which was part of our
00:462015 academic roundtable to offer
00:49another perspective Lee's company
00:51tachyon Nexus came out of work he did as
00:54a student in the amp Lev and the
00:56resulting open source project Lee
00:58describes the struggles and victories in
01:00making the transition from student to
01:02founder and leader of a company the
01:05ampulla begues ample and what every
01:07academic and entrepreneur can learn from
01:09it on this segment of the a 16 z podcast
01:13so let's start with you Michael
01:15you were entrepreneur in between being
01:18an academic for a long time but amp lat
01:20come out of the desire to do what and I
01:26mean and and how did you sort of make
01:28yeah so where do you start
01:32so yeah so it was interesting because I
01:34was off I took two years off from
01:37teaching at Berkeley to get a company
01:40started and then you know kind of came
01:42back after that Yan Stojko who's also
01:44one of the founders of the amp lab at
01:47the same exact time went off and was way
01:50for two years doing a company and when
01:52we came back I think there were a couple
01:55things going on one is after you've done
01:58a start-up academia can seem a little
02:04slow quiet and so you know you get used
02:08to being in this environment where kind
02:09of every day is is kind of a life or
02:11death situation right and you walk in
02:14and you know mostly in academia there
02:17it's definitely life and death
02:18situations but they're sort of not every
02:20day and so I think you know I won't
02:24speak for yawn but I know from my
02:26perspective I was kind of craving a
02:28little bit more excitement a little bit
02:30a little bit more interaction in a lot
02:32of traditional academic environments or
02:34a little solitary and so there was a
02:38project going on at Berkeley called the
02:40rad lab which was a project of systems
02:45people and machine learning people who
02:46had gotten together to do what at the
02:48time was called autonomic computing
02:50where you use machine learning
02:52algorithms to help manage large computer
02:54systems and so I ended up joining that
02:58thing not because I was so interested
03:00not out of a commute computing but just
03:02because I like the environment of having
03:03all those people around the problem was
03:07that rad lab was getting ready to wind
03:09down it had been started with a
03:10five-year timeline and so they were
03:13thinking of splitting up they were
03:14talking about how to split up the lab
03:16you know we had this big open
03:17collaborative space and you know when I
03:20had been out my company I saw what now
03:25is what's called Big Data revolution or
03:29whatever but every company I talked to
03:30to try to sell them stuff it was just
03:32clear they were getting more and more
03:33data and just that seemed to me to be
03:36kind of the perfect problem for this
03:38group of people who on one hand did
03:40large-scale computing on the other hand
03:42big machine learning and started talking
03:44to you on who you know was out seeing
03:46similar things and we realized that was
03:49so it was it was a technology trend that
03:52you really I mean it's sort of bigger
03:54than that in a lot of ways but it goes
03:55out from there but big data was it yeah
03:58yeah and and and the I think just about
04:01everyone here's a professor I mean the
04:03the message I give so when we we have a
04:05new chancellor at Berkeley and when he
04:07showed up he had heard about the Apple a
04:08band he kind of wanted to understand
04:10more about it so he invited me to tell
04:12him about it and and the thing I said to
04:13him was you know if you remember one
04:15thing from this conversation
04:16here's what I'd like it to be is that
04:20because there were two faculty members
04:22who got to go off away from the
04:25University for you know a couple of
04:27years and engage with what was going on
04:29out in the world and that's where we saw
04:31it coming and we saw it before a lot of
04:33other people in academia because of it
04:35so amp lab you know it's working I mean
04:38it's from our perspective in Peter I
04:40want you to jump in here it's like it's
04:41working incredibly well both from an
04:42academic standpoint you know to get this
04:45kind of critical mass of people and
04:47ideas together and then or about the
04:48other end comes you know all these great
04:51companies and ideas and projects but so
04:55and I want you both to an assessment
04:56Peter you first what are the ingredients
04:57sort of the amp lab has mixed up or that
05:03you can identify that makes amp lab work
05:05so well from from your perspective from
05:09my perspective I think the the most the
05:13most interesting aspect of the amp lev
05:16is sort of this macro trend in you know
05:20system software now being done by a
05:23whole new generation of folks enterprise
05:27and infrastructure software system
05:29software was traditionally done by folks
05:32let's say leaving Cisco or leaving
05:34Oracle and it was it's always was a
05:37derivative like you work at Cisco and
05:40then you go do a start-up but your work
05:41at Oracle and go do a start-up and it
05:43was always kind of to me there was a bit
05:45of incrementalism and it was sort of an
05:49older an older generation of people
05:51actually doing the work on those some of
05:54those new companies and what's happened
05:56I you know in many universities but amp
05:59leavened specific as we find that
06:02there's a there's now the new generation
06:04working on the most interesting computer
06:07science problems applied to system
06:09software and yes big data but its
06:12database is operating system software
06:14kind of all the infrastructure pieces
06:15and I got to tell you that has
06:19fundamentally moved the needle in the
06:21industry to where you know the companies
06:24coming out of there are way way
06:27different than what I
06:28found you know from people coming out of
06:30let's say Oracle or Cisco it's just a
06:32revolutionary new thinking and kind of
06:36it just opens your eyes to really new
06:37ways of doing computing that's been that
06:40to me that's been amazing if you were to
06:43distill that to an ingredient for lack
06:46I mean is it focus then is it like that
06:48you guys don't try and boil the ocean
06:49and solve every poem but you're focused
06:51on look if I want to do research and do
06:54my graduate work in infrastructure and
06:56you know you big data
06:58I go to amplify mean I think from from
07:13my perspective I think another way to
07:15say what Peter was saying was first of
07:17all open source has made a lot of this
07:20possible and so yeah you know I always
07:23tell people when I started my career as
07:25a database systems researcher you know
07:28you'd come up with an idea if you
07:30thought was a pretty good idea you you
07:31know you go and you tell you'd give a
07:33talk at IBM and you'd give a talk at
07:35Oracle and you give it at Microsoft you
07:37know and you know either they would take
07:40your idea and steal right and then you
07:47know maybe you'd hear about it maybe you
07:49wouldn't are they'd say oh no we don't
07:51where is now you know the way our
07:52students do work is you know they they
07:55work on a piece of software if it looks
07:57useful they log about it or somebody
08:00else does or they put it out in the am
08:01flap repo and flip a switch and all
08:04sudden people start trying you know so
08:05that the the friction from you know
08:08student idea to actually people using it
08:11I think that's just right it was a great
08:12the friction has been weight reduced and
08:15so then that friction and that
08:17relationship between the corporate world
08:19then how do they arrive at the amp lab
08:21and then how do you leverage them so
08:24yeah so I mean I don't know if anyone
08:26here knows a lot about the implied one
08:28of the things that's interesting from a
08:30academic perspective and the end up is
08:32our funding comes about half from you
08:34know the government from NSF and DARPA
08:35in places like that and have fun half
08:38from companies so we currently have 30
08:41that helps sponsor the work and we
08:45engaged with them very early telling
08:49them not only what we've done but what
08:50we're thinking is doing and that has a
08:52couple of nice nice features one is we
08:55get kind of instant feedback on what
08:57sounds right to them and what what
08:59doesn't ring true and we also get
09:02feedback from them on you know problems
09:05that they're having that we're not
09:06really looking at and then the other
09:08advantage it has is that it by getting
09:10them involved early they they're more
09:13inclined to to try to you know use what
09:15we do and give us feedback right so who
09:19gets IP rights among all those
09:21contributing companies brilliant
09:23question so and and this was done way
09:27before amp lab you know Berkeley has a
09:30long history of open source and so
09:32there's been a bunch of projects that
09:33work this way and the way it works is we
09:38don't give the companies any IP rights
09:40and we work for the companies do we
09:49could talk about who it might not but
09:51but you know the idea is we do
09:54everything an open source and and so
09:57then the next question is well then
09:59given that the companies could get at
10:00the stuff anyway why do they give us
10:02money and and that's a more complicated
10:04question different companies have
10:06different reasons well Peter you have a
10:08good answer for you invest in you know
10:10and multiple open source companies so or
10:13open service projects that become
10:14companies in a different sort and we'll
10:16talk H Y with you about it more in a
10:19minute but how does it work for you and
10:22when you're approaching an academic
10:24institution where do you start and you
10:27know let's take amp lab like how did
10:28that relationship even blossom well we
10:31had a number of connections in to
10:33amplify ike and i will actually work
10:34together in a past generation so we knew
10:38knew each other Ben of course knew folks
10:40there and so we kind of look at it's
10:42always been a matter of finding the you
10:44know a couple of people inside a
10:46university and then and then yeah
10:50building on that relationship I think
10:52what's been unique about Amp lab it's
10:53actually one relationship with multiple
10:56projects typically it's one relationship
10:58with one project so it'd be has become a
11:00very effective sort of mechanism for us
11:03to go in and actually look at all the
11:05projects going on it's just a
11:07centralized place at Berkeley which has
11:09been fantastic and that wok I mean once
11:12we see the projects and we look at them
11:15and evaluate them as potential
11:16businesses we invest as we would any
11:19other company right and so then it
11:22starts as you know commercial enterprise
11:23and we go from there
11:26maybe you can respond to this question
11:28of IP in and the corporate world because
11:30you know again in the open source
11:33project that become companies you have a
11:35very specific kind of view of the world
11:38and why open source works as a business
11:40and that might sort of answer up to why
11:42the corporate world is fine with you
11:44know investing the way they do well one
11:48of my beliefs about open source which
11:50gets to sort of this point the IP
11:53ownership first of all with open source
11:56IP ownership is to me less important
11:59than who actually wrote the code and so
12:01what I look at fundamentally for an open
12:04source project is I'd like to see that
12:08the inventors of the code are part of
12:10the company and whenever you have a fork
12:12which is you know a derivative to
12:16projects that use the same code base
12:17after all it's open source the fork
12:21typically is not as thoughtful as the
12:25folks who have invented it and all the
12:27projects coming out of amp lab in fact
12:29one of the rules I mean I do most of our
12:31open source investing and that's one of
12:33my guidepost is to say look it's got to
12:35be a real exception if we're going to
12:37invest in something that's a fork of the
12:39real guy the real folks doing it and so
12:42what we have found out of amp lat it is
12:44the inventors of the particular project
12:47who are starting the company and that
12:49gets you great competitive
12:51differentiation because if anyone Forks
12:54it and tries to come up with another
12:56team it's like look we're the inventors
12:57of the of the project we run the roadmap
13:00if you want support or add-on
13:02or whatever it might be the the the
13:06floor company is never going to be as
13:08strong as the one who has invented it so
13:11the project's coming out of amp lab
13:14certainly fill that first requirement
13:16for me you know it's an interesting
13:20question that we get periodically
13:23whenever somebody new shows up in the
13:26administration at Berkeley of you know
13:28why are we doing this open source stuff
13:30why don't we patent these things and
13:32then we there's this conversation that
13:34you have to go through about well okay
13:35you know how many software patents you
13:37know we brought in you know eight
13:40figures worth of you know industrial
13:43donations for amp lab software patents
13:45has the University of California ever
13:47had that brought in that much money and
13:49the answer is maybe one they had a
13:54patent on the browser I believe right
13:59yeah Marx out here but one of the very
14:02early browsers was that at Berkeley
14:04let's put it that way and they had some
14:05patents on that and I think that was the
14:08one that they made money on but I'm not
14:09even under percent sure and and the good
14:11news is it'd be hard to find an
14:12administrator at Berkeley who could give
14:13you the answer either and so that's the
14:15thing is once once you you know you have
14:18to make this argument about the
14:19importance of you know building the
14:21brand building the the the level of
14:24activity and then you know it's a little
14:26indirect but yeah these you know the
14:29opportunities for philanthropy start
14:31coming in and in fact Berkeley now
14:32started this thing they call the
14:34founders pledge where they ask people
14:36who are doing things in the university
14:39you know the pledge you know some
14:42portion of their future revenues or
14:45profits or whatever and it's non-binding
14:49but whatever so they're working on it
14:51Michael how do you view projects that
14:55come out of amp have been research yeah
14:56is there in a sort of a sense that you
14:59have like oh my god here's another one
15:01like this one's gonna be off to the
15:02races and and let's go or how does I
15:08just wonder like the sort of when
15:09research becomes commercialized or
15:12commercializable what are the clues
15:17I don't know about commercial or
15:19commercializable she sets a that's a
15:23tougher one I mean just I thought where
15:24you were going with that was sort of
15:25what's a good idea and what's not a good
15:26idea well we can go there whether it's
15:29from from a reason or or do you guys
15:30even care whether it's yeah because
15:36that's not really high on our list when
15:37right when we start thinking about what
15:39to do yeah that's what I would imagine
15:41yeah yeah so what's a good idea and then
15:43I'll ask you the commercializable part
15:45because you probably by the way you know
15:50I don't want to say we don't care I'm
15:52just saying that's not the first right
15:53thing but but I think you know in a lot
15:59of it just has to do with somebody in
16:01the lab being you know being I mean it's
16:03gonna sound cliche I guess but being
16:05passionate or being excited about
16:07working on something because because you
16:09know that's what happens we you know a
16:10lot of our most successful projects now
16:11are things where you know somebody goes
16:14off and and and works on some piece of
16:16the system you know puts it out on
16:19github and somebody finds it and all of
16:21a sudden there's a community right
16:23starting to form around it and so you
16:26know we're in this environment now where
16:27ideas go out there quickly and then very
16:30quickly you sort of see what gets
16:31traction and what doesn't so its
16:33traction it's sort of passion it's you
16:36know and it's gonna be good idea I mean
16:37that's what a good idea is traction and
16:39passion it absolutely and well and also
16:41I mean we are University we're we're a
16:44research lab where our main output is
16:47great students and you know one thing I
16:51want to mention too to all the
16:53professors here since you know you might
16:55think well okay but you know I have this
16:57problem I have to decide do I want to do
16:59you know good research or do I want to
17:01do things that are you know maybe useful
17:04right and you know if I think most of
17:07people since you're sitting here I'm
17:08probably preaching to the choir but
17:10that's a false dichotomy I mean we were
17:13very fortunate left in this current year
17:16you know you all know the ACM
17:19dissertation Awards they give an award
17:21to sort of the the the the best PhD
17:25worked in computer science around the
17:28so every school gets nominate two people
17:31mm-hmm I'm chair of the department it
17:33just so happened last year the two
17:34people got nominated were from a up
17:35laughs just a coincidence and you know
17:41one of Matei Zarya won the award he's
17:44guided spark John du chí who was the
17:46other guy who nominated we've got one of
17:48the two honorable mentions so of the
17:50three awards ACM gave worldwide last
17:52year two of them came from an play well
17:54what does that say about technology and
17:56kind of its place in just broader
17:59society then and and culture and
18:01research etc in academia I mean I think
18:04there's a probably better premonition
18:07now for impact right and maybe there was
18:10an academia you know we might arrive
18:12years ago I think that's right if you
18:14look at you know a lot of innovation up
18:18till well to miri up until recently came
18:21from like Bell Labs in IBM and Microsoft
18:23Research and that's where a lot of the
18:26new I would say the new commercial ideas
18:28came from folks leaving those or pushing
18:31the envelope now I see a lot of it
18:33coming out of universities which is that
18:35that's fantastic I have to tell this
18:38great story though if you don't mind
18:39just please so we had a delegation from
18:41China visiting this weekend and we
18:45we showed them you know the amp lab all
18:48the stuff we were doing to show them a
18:49bunch of other data science things going
18:50on around Berkeley and at the end the
18:52president of this university was very
18:54prestigious university in China said to
18:56us well whose plan was this you know who
19:01came up this is you know the work is
19:03wonderful who came up with this plan and
19:06you know we all sort of just looked each
19:07other and because they really wasn't a
19:10planning it's just it was really just a
19:12bunch of you know different faculty who
19:14were doing what they were interested in
19:16and and decided to collaborate let's
19:19share some gears a little bit to
19:21advances in computer science what are
19:23the most important things you're
19:25starting to see and what it what are
19:27getting your students excited and Peter
19:28you know answer this question is as well
19:31I think so the whole idea Big Data has
19:35multiple I think it's a generational
19:37kind of there's multiple aspects of big
19:40now that we're able to collect data and
19:43now we're able to do real-time big data
19:46I think machine learning and deep
19:48learning are probably the next I mean to
19:50me I see a lot of those projects you
19:52know people coming either at a
19:54university out of Google out of a
19:56variety of places to think about how we
19:59process and become much more predictive
20:02about the information that we have and
20:04I'm actually pretty excited about that I
20:07mean there's a lot of noise in the space
20:08right now there's a lot of people sort
20:11of interested in looking at that but I
20:13think there's a relatively new frontier
20:16that's occurring at a lot of
20:17universities around sort of machine
20:19learning and and you know deep deep
20:22learning yeah I mean so machine learning
20:24deep learning I also want to know like
20:25with new databases would be another
20:28thing like that anyway with your
20:31colleagues outside of this yes
20:33Department is that all part of their you
20:35know language now like oh I'm working on
20:37a big you know amp lab we're all about
20:38big data machine learning you know deep
20:40learning did they get that yeah I think
20:43a lot of people do I mean the one of the
20:46the big buzz words around our campus
20:49these days I think most campuses it's
20:50data science and and that one tends to
20:53be a little more inclusive and big data
20:55big data is looked a little bit as kind
20:57of a computer science thing right
20:58whereas data science really all around
21:01campus I mean even astronomers go around
21:04saying that their data scientists for
21:06example so so so so data Sciences is a
21:14really hot topic and and one you know
21:18about how do you solve real analytical
21:20problems a bunch of domains I've been
21:24getting interested in in things around
21:27trying to I've been doing databases my
21:30whole career so I'm sort of interested
21:32in reaching out from from the database
21:35to actually go out and affect the world
21:37so you know if you look at things like
21:40cloud robotics or you know think about
21:44things you can do with with with with
21:46drones or anything like that I think a
21:48lot of people are getting excited about
21:50Internet of Things but in the new
21:53in the old version of it and other
21:55things it was just you got sensors out
21:56there and he collecting the data and
21:57you're doing something with it
21:58and the new version of Internet of
22:00Things are actually going out and
22:01interacting with the world and that
22:03involves also the people that are in the
22:05world that have to coexist with all
22:07these machines one of the areas that
22:09we're seeing from you know both from
22:12universities and new projects coming up
22:14is this whole idea of moving the compute
22:17out to the end point so like hi it's
22:19really interesting to see this the ebon
22:21flow of computing from sort of
22:23centralized to distributed we've been in
22:26this centralized world cloud computing
22:27and then you have the endpoint which
22:29shows pixels and now we're starting to
22:31see a number of really interesting
22:34projects where the endpoint our
22:36supercomputer in our hand is actually
22:37being used as a computer not a display
22:40device and I think the Internet of
22:42Things and drones and however you want
22:44to define these this sort of range of
22:47computers that exist not so much in our
22:49pocket but out there in the world I
22:50believe that there's going to be
22:52computing that will be fundamental to
22:54how we start to think about data and you
22:58know real-time analytics that occur not
23:00so much in the cloud but at the endpoint
23:02and so we're starting to see that in
23:04fact out of universities and all of that
23:06so there is a trend now back towards I
23:08predicted distributed computing will be
23:11back and it's going to be with all these
23:13endpoint devices and drugs Michael I've
23:16been ask you one last question will
23:17brain hy amp the P is for people you
23:21just mentioned people you know and maybe
23:23you guys are a little bit ahead of
23:24yourselves five years ago but how does
23:27the how do people fit into sort of the
23:29amp you of the world and and again when
23:32you think about you know these sort of
23:34database backed internet of things and I
23:37have you bringing all the people and all
23:38the data how do people fit into it all
23:41yeah it's interesting because the the
23:44the view of people in the lab is or the
23:47role of people in the lab has evolved
23:49over the since we started when we
23:52initially started the idea that we had
23:54was that algorithms in terms of you know
23:57machine learning you know fistic 'el
24:00processing and machines the idea was
24:01cloud computing and large clusters and
24:06we were thinking about well these are
24:08the three resources that you have
24:09available three types of resources you
24:11have available to make sense of data mmm
24:13makes us happy data in particular so our
24:15view of people when we originally
24:17started was to think about human
24:19computation and crowdsourcing you have
24:21in fact Tim sitting back there one of
24:24the Tim was one of the early people in
24:26the up lab and we worked on a project
24:29where it's called crowd DB where you
24:34know I like to call it the world's
24:35stupidest database system where you know
24:37if you asked it a question and you
24:39weren't attached to the network it would
24:41just say gee I don't know they answered
24:43that question what do you think would
24:49create a bunch of jobs of Mechanical
24:51Turk or something and send them out and
24:53let the crowd answer and then also it
24:55wasn't such a dumb database system
24:56anymore and so you know we still have a
24:59big part of what we're doing in terms of
25:00bringing in human-in-the-loop analytics
25:03looking at how do you get people either
25:05as you know individual experts or
25:08analysts or as or as crowds to do data
25:10cleaning the idea was you wanted to be
25:12able to bring in people to solve those
25:14parts of these you know machine learning
25:16problems that the machine learning
25:17wasn't quite up front of the speed for
25:19right and how things have evolved kind
25:21of over the five years is now we're also
25:23being a little more concerned with the
25:25fact that you know ultimately the
25:27results of the analysis is in many cases
25:29going to end up in front of a person so
25:31how do you support that person is a job
25:33but the original vision was really about
25:35human computation and crowdsourcing
25:36great well Michael we're gonna kick you
25:39off now and so H whether he was the
25:49co-creator of tachyon right and so he is
25:52as you say Peter the original og open
25:55source guy on this project and you can
25:58describe it it's a memory centric
26:00distributed storage system which you've
26:02now formed tachyon Nexus which is the
26:06company that is built around that maybe
26:08just give us a brief discussion of
26:09tachyon and and what it is that that the
26:12problem that you've gone after and
26:15actually why why you wanted to address
26:20oh okay so so Taikan is the memory
26:24century should be in storage and I think
26:26from from the high level the vision is
26:29that memory centric memory centric
26:31computing is the future and you have
26:32computer of his Jorge there and we want
26:34to be the storage layer and and why I
26:38wanted to do this problem it's a that's
26:41a interesting so so personally I'm very
26:45interested in storage but why joint like
26:47Berkeley and Platt like four years ago
26:53we have other two projects like mesos
26:56Apache missiles and the party spark both
26:59are very successful and actually the
27:02company handles product also like
27:03founded by ax and Jason harus and and
27:07but at the time the storage piece is
27:10still missing so so I was fortunate that
27:13my advisors and the lab give me the
27:16opportunity to work on this piece and
27:18it's just a very fascinating to work on
27:22something you think is a very important
27:25interesting yeah so Peter let's get back
27:27to this good idea bad idea question you
27:29know clearly what what hy was working on
27:32you thought was a good idea so what were
27:34the attributes of that idea that you
27:36thought then and or evidence that you
27:39thought this was something that you know
27:41you wanted to put your time and effort
27:42and and the firm's unit behind when I
27:45when we as a firm look at you know
27:48what's a good idea versus maybe not such
27:50a good idea and time will tell you never
27:52know but we go into it all our
27:55investments we obviously go into it
27:56think is a freaking great idea they
27:59don't open it out that way so that's
28:00just one caveat the view here I mean I
28:05have been in the storage space for a
28:07long time and when we met one of the one
28:12of the concepts and we've been talking
28:13about internally imagine if memory in a
28:15computer flatten down to where there was
28:17only one memory hierarchy or let's say -
28:19ok I know you may not think so but like
28:22well hang on no spinning disks are going
28:25away ok so if you take that to the next
28:29lot you know we have SSDs and we have
28:32you know CPU memory and part of my
28:36belief was that if memory gets cheaper
28:39and cheaper why don't we have a file
28:41system and why don't we have an
28:43architecture that actually supports this
28:46new type of computing because lots of
28:50community I mean computers and operating
28:52system code and the way we write files
28:54and the way we treat structures and all
28:56this stuff has all been written under
28:58the assumption that there's is there's
28:59this memory hierarchy that goes from
29:02very fast and expensive too cheap and
29:05slow and everything we do bounces
29:07through that hierarchy and my belief
29:09whether you believe you know look I love
29:11that people disagree you know for the
29:13most part when more people disagree
29:15that's the project we want to go invest
29:17in because like you know it's it's too
29:20crazy for you know hell that'll never
29:22work but when you know when enough
29:24people disagree that there will never be
29:26flat memory structures then it's a great
29:29idea you know at least to get started
29:30right it may never happen but that was a
29:32thesis that memory was going to get much
29:35I mean look you look at cell phones and
29:38you look at the memory and there may not
29:39be fault tolerant and main you know fall
29:42apart all the time but part of the
29:43systems we want to go solve for that but
29:46it's very inexpensive and so to the
29:47extent that the mobile supply chain
29:49starts to eat the backend data center
29:51was kind of our thinking on how this
29:54memory you know in in in core memory
29:57starts to start and starts to flatten
29:59itself so with that if you make that
30:01assumption and if you believe it which I
30:05you know believe to some degree then
30:08this project exactly maps into that into
30:11that future and it and even I would say
30:13even if it doesn't quite map exactly
30:15like there it's still a great project so
30:18like the the grand the grand view would
30:21be memory flattens and this becomes the
30:23basically the in-memory file system for
30:26all computing and if we don't quite get
30:28there it's still a great file system for
30:30big data and other applications so
30:32that's sort of the you know we sort of
30:34look at the Grand View and then say ok
30:36well if that doesn't happen because we
30:39were wrong about that what else can this
30:41become so so that I mean like when we
30:46look at ideas on that dimension to say
30:49you know what what what transformations
30:52might occur in the future to where there
30:54really is a big need for a whole new
30:56design of something and this was one of
30:59those ideas now I do come like I do
31:02build taxonomy x' in my brain and
31:05probably the simplest thing that I do
31:08and I'll leave you with this like how do
31:10I actually like when you see something
31:11how do you know or how do you choose to
31:13invest in it so what I have I've been
31:15here now five years and my my whole
31:17philosophy around this is I take
31:20something that's very popular right now
31:22cloud computing and I say what happens
31:26when cloud computing doesn't exist
31:27anymore what fills that place and all of
31:31a sudden like and I and I put in my head
31:34the most ludicrous things because 15
31:36years ago if you assume Microsoft might
31:38not be at the top of the compute gene
31:40and you know everyone would look at you
31:42like you were crazy or digital equipment
31:44or you know you say Google now Oh Google
31:47you know may not be at the top of the
31:48food chain in the future and everyone
31:50would look at you like well you're crazy
31:51they'll be there forever so I think by
31:54take for me taking away something that's
31:57out there it has been a really
31:59interesting exercise and then you find
32:01projects that actually you know I don't
32:04go out and look to say what happens
32:06let's say when VMware goes away or
32:08whatever you sort of wait I have these
32:10views in my mind and then all of a
32:12sudden you start to see things that fill
32:14the spaces and so that's kind of my you
32:17know some of my models and a lot of
32:19actually a lot of our amp lab
32:21investments have been based on look
32:24phenomenal entrepreneurs great ideas in
32:26their own right but then me also
32:29thinking okay well like if something
32:31isn't there maybe this is the thing that
32:32fills it is that part of the sort of
32:35ecosystem an amp lab dough tube because
32:37it's it kind of it's hitting all those
32:40well amp lab certainly if you look at
32:43spark maces and tachyon I should say
32:46tachyon spark maysa in that sort of
32:49architectural order it does create a
32:52full stack of you know kind of the next
32:55generation of big data kind of
32:57infrastructure and that's really I'm for
33:00I mean it's in memory it's got like blah
33:02blah you know scale out all this stuff
33:03that currently doesn't really exist
33:08and so yeah we've you know kind of and
33:11it's part of them each way do you think
33:13amp lab is a place like that because
33:14there's this kind of you know ecosystem
33:17of smart people working on these
33:19different parts of it so look somebody
33:20else is already here so I'm gonna go
33:22here I mean and I mean does it is it
33:26this sort of you know ecosystem that
33:28exists there and you need to figure out
33:30where you fit within that ecosystem or
33:32is it just like here's my interest in
33:33I'm gonna go after it but from my
33:36perspective I think they're lab give us
33:39a very great environment and from the
33:42individual perspective we don't have
33:43that much pressure like to you know in
33:47terms of paper publishing those type of
33:49things right and and in the meantime in
33:52the meantime we can we can focus a to
33:55for a pretty long time and you know
33:58different people yeah but their only
34:00interest we have a pretty large lab and
34:02we have a lot of people to bounce the
34:04idea we can talk with professors and
34:06talk to students actually the professor
34:09they don't have an office like they just
34:10seem in a city in the same cube as we
34:12students there right so it's it's a very
34:15good environment I think we can we can
34:17leverage in the meantime the lab also
34:20talk with the industry very regularly
34:22like we have the meter retreat thing
34:25like twice a year and we have like maybe
34:2720 industry sponsors they came to we
34:31talk communicate I think those are very
34:33helpful Peter in in your experience what
34:37are some of the traits that you see in
34:38hy and other entrepreneurs that
34:41correlate with success one is just look
34:46a deep understanding of the space that
34:48they're in that's and the passion to
34:51we always want to see that I think also
34:54it's the it's the willingness to learn
34:57all the things that the entrepreneur
34:59doesn't yet know and so look I mean you
35:04know hy is a great example and a very
35:06positive example he's very passionate he
35:08understands his area better than anyone
35:12but there's a lot of other things that
35:15you need to do to learn how to run a
35:17company and all the little you know the
35:19little things that you all might think
35:20are easy and stupid actually trip you up
35:23and so what we want to look for is
35:26somebody who is going to be coachable
35:29because there is sort of a pattern
35:31there's there's a blueprint on how to go
35:33do this I mean not now that I've done
35:35this multiple times and pretty much take
35:37somebody through here's what you do this
35:39week next week next month in terms of
35:41building out a sales organization in
35:43terms of hiring a product manager in
35:45terms of hiring a CFO like there's all
35:48marketing how do you do that I mean
35:50there's all these other pieces to
35:51building a company besides the
35:53technology and so one of the attributes
35:55that we want to try to assess upfront is
36:00is the entrepreneur and you know PhD
36:03student grad capable of being coached
36:07into these new areas because let's face
36:08it I mean a lot of folks don't want to
36:11be coached it's like look I'm just gonna
36:12go do my and like leave me alone I
36:15don't want to learn about this other
36:16stuff and you probably can't build a
36:18company if the only thing you do is
36:21focus on your technology and you can so
36:26that's what I would say you thought you
36:27never would have to deal with and but
36:29you either love or somebody else would
36:31do it I mean I just I think I just wanna
36:37have a goal and I want to I just want to
36:41achieve the goal and and along the way
36:44there are different things for example
36:45like Peter mentioned in terms of hiring
36:47like process it's many many things in
36:50the hiring like at the beginning you
36:52know we don't have a process right and
36:54then you have a process and you multiply
36:56the process improve the process in the
36:57meantime you have this and that
36:59condition and we talk about this type of
37:01stuff and some some cases very is that
37:04it's cuz it's it's very new to me right
37:06it's very interesting it's also very
37:07interesting to me is there a class or
37:09something in school you wished you had
37:11learned that would have helped you you
37:13take in more marketing classes or any
37:14for that matter um you know like I I
37:19think I think school taught me is really
37:21a lot and and that's the reason you know
37:24like I can still do some stuff
37:26so but in the meantime like I also think
37:29a lot of things you should like you can
37:31learn a lot away I see as you go if I
37:35have to put something I'm an engineer
37:38background in the sea-ice background so
37:39back to school maybe from the technical
37:42perspective is like you know be great if
37:45more students know how to build a
37:47production level like system if they
37:50want to do a system level thank right
37:52for other stuff I think I mean you can
37:54learn along the way if if you're willing
37:57to learn yeah very great yeah maybe
38:00you're not quite there yet but your
38:01views on the other parts of the business
38:03have they changed you know so sales
38:06marketing you know all that stuff are
38:08you just learning as you go and are you
38:09a bottleneck still are you so so in tune
38:14to not being a bottleneck that that
38:16that's not happening I try I try not to
38:20be the button that yeah but it's like in
38:22real I mean you got a lot of traffic
38:25right so I mean Peter taught me a lot
38:29like you have to prioritize so in some
38:32sense we are always about to that since
38:34you can you have infinite things to do
38:36but in the meantime like we have we've
38:39got a great team so I think our people
38:43are great and they can do things like
38:44like move things forward fast and and
38:48without like my my my interaction there
38:53right so I'm pretty happy about that
38:54part I think one of the most difficult
38:56transitions a person makes in their
38:59career whether it's from academia or
39:01from just an individual is from
39:04individual contributor to being a
39:05manager and leader and when you're an
39:09individual contributor you do everything
39:11and you're the best at it and you hack
39:13your code and you do whatever and when
39:16you become a manager you have to do that
39:18through other people and that transition
39:20is non-intuitive and very very difficult
39:23for I mean I had to go through it - I'm
39:26not I mean yes of course I went to
39:28college and all that I'm not at your
39:30level but it didn't matter I went from
39:32individual computer programmer to a
39:35manager and it's incredibly difficult
39:37because most of your time look
39:40as an individual contributor your time
39:42is spent writing code and and when you
39:45become a leader which is the nicer way
39:49of saying manager a leader most of your
39:52time really ought to be spent in hiring
39:54and coaching new people to come on and
39:57that's completely counterintuitive it's
39:59like look if I write more code this week
40:02the project's gonna move ahead but if
40:04you don't hire anybody you're gonna
40:05continue to be the bottleneck and I'm
40:07like that transition and you know you're
40:10still going through it that transition
40:12is a writ it's difficult for everyone
40:14and I think particularly difficult for
40:16people who have written the code and
40:18like you know every line and every
40:20comment and all that stuff it's
40:22particularly difficult to go through
40:24that it's why what hurdle did you
40:27recently put behind you and what hurdle
40:30is next I mean so as I already start up
40:33like like one challenges you always want
40:36to get the best people and takes time
40:37and especially our part is growing
40:40pretty fast so we all have a lot of
40:43inbound traffic we always have trouble
40:45like in fully exploit that so and what's
40:50next what's the hurdle that you're
40:51looking ahead is it still hiring I mean
40:54it would just keep hard hiring you know
40:58I think I mean I'll you know we spend a
41:01lot of time on this I think the next
41:03hurdle for these guys is to figure out
41:04how do you commercialize the open source
41:07project that you have it gets to what
41:10exactly is the business model what's
41:12paid for what's free is anything free
41:14like coming up with that business model
41:16such because if you don't have that
41:20early what starts and you don't know
41:23then potential customers and partners
41:25don't know what you're going to charge
41:27for in the future and all that and it
41:29makes it the longer you go on without
41:31having a model the harder it is to roll
41:33back some of those things I like in a
41:35year from now so go well we're gonna
41:36charge for that and everyone's like well
41:38we've been using it for free for a year
41:40like you can't charge for that anymore
41:42and so it being able to lead with the
41:45crumbs along the way and saying okay
41:47here's what we're going to be doing get
41:50ready for this kind of gets the Commun
41:54you know precondition to kind of
41:56understand what the model is going to be
41:58so I think right now that's what we're
42:01kind of talking about age my last
42:04question for you you were back at that
42:06point where you were the bottleneck you
42:07knew it you had a choice to not go ahead
42:11and and start tacking Nexus mm-hmm do
42:15you still think you made the right
42:15choice absolutely yeah I think I made
42:19yeah wait you know in five years Michael
42:24berkland H Reilly Peter Levine thank you