00:00hello I'm Peter Levine here at
00:02Andreessen Horowitz and I'm with your
00:05Nova who is the CEO of max 'te and with
00:08Flo Lee bird who's the CEO of Mesa
00:11sphere great to have you guys thanks for
00:15thanks for joining it's a pleasure
00:17yeah thank you so I wanted to spend a
00:20little bit of time talking about the the
00:23consumerization of IT and by
00:25consumerization I don't mean bringing
00:28your own device you know mobile device
00:31to work like that's happening and that's
00:35already done with what I'm specifically
00:37talking about with respect to the
00:39consumerization of IT is the idea that
00:42the components of mobile devices will
00:47become the basis for hardware in the
00:50data center of the future and so not
00:52unlike we saw the PC revolution replace
00:56mainframes in the data center the belief
01:00I have this strong belief that the
01:03innovations in mobile hardware over the
01:05past 5 10 years will replace sort of the
01:09PC architectures in the data center to
01:12where the mobile supply chain
01:14effectively eats the hardware of the PC
01:18world in in the data center and you can
01:20imagine you know a billion mobile
01:23devices being the basis of the hardware
01:25infrastructure and then of course you
01:28know the belief is that then taking all
01:30that hardware and putting it together
01:32requires some level of sophisticated
01:36software that then would aggregate and
01:39bring kind of the enterprise class
01:42capabilities out of this underlying
01:45commodity hardware so what do you think
01:48of that like you know how have you
01:50thought about this in your your own
01:51business maybe start with flow sure so I
01:54think Google in the past has already
01:56shown that they're using commodity
01:58off-the-shelf components in order to
02:00built their entire data center
02:01infrastructure and I think with open
02:03compute at Facebook we're seeing a
02:05similar trend across the industry now
02:07where everybody is using commodity
02:09components and of course that means that
02:12there are more failures
02:13so software needs to be able to handle
02:16those failures we're going to see
02:18plug-and-play data centers where we can
02:20just add chips remove chips and they're
02:23just added to the pool I think that's
02:25certainly a trend yes I I think that we
02:29see basically the similar trend in the
02:33industry so if you go back to the past
02:36storage was originally proprietary
02:38system that were custom build and all
02:41the time people started to use commodity
02:44hardware to build new storage system and
02:46the value moved to software and those
02:49are the x86 standard PC hardware and
02:52there is no reason why the next
02:55generation of hardware platform which is
02:58basically our processor and stuff for
03:00like flesh coming from the the consumer
03:05side will not take over because anytime
03:08communities commoditization happened in
03:11the hardware from a correspondent of you
03:13for me from a advanced technology point
03:16of view it's always winning and what you
03:19need to make it all work together is
03:21smart software that put all the pieces
03:23together and make relatively unreliable
03:26Hardware work in a better way than it
03:30would work without this sophisticated
03:32software so do you guys you know we talk
03:36about arm and flash is kind of principle
03:38there is no spinning disk in a mobile
03:40phone there's you know most mobile
03:43phones don't have intel processors they
03:46have different processors do you really
03:48think that these components like become
03:50the basis of the data center and you
03:52know if I'm if I'm a existing company in
03:55the data center space I'm like saying
03:57well that stuff will never work like you
03:59know we can't expect commodity an ARM
04:02chip really to power the data center or
04:05for commodity flash that's gonna fail
04:07all the time do you guys like what's
04:09driving that trend and do you really
04:11think it's gonna happen that's it yeah I
04:13think it's definitely going going to
04:15take place and and you know the reasons
04:17if you look at the challenges in the
04:19data center today there are around power
04:22and cooling and arm by its nature
04:25provide a better food
04:27so it is true that today ARM processor
04:30is not as powerful as an x86 but as the
04:34PC overall in majority of cases replace
04:39the mainframe because you can always
04:41have a bunch of PC that are performing
04:43as good as a mainframe the same thing I
04:47believe will happen with processors well
04:50ARM processor which is much more or much
04:53better in power and cooling over time
04:55will become powerful enough so that's on
04:57the side of the processor flesh is
05:00definitely going to take over again
05:02power cooling and especially performance
05:04more flesh even if reliability is not
05:07there and software is going to bridge
05:10the gap and basically provide the
05:14performance that you need with the
05:16reliability that you need and overcome
05:18the challenges that you have today with
05:20commodity flash which is still not as
05:22reliable as disk drive but the software
05:25and the communitisation of hardware and
05:29the lower cost will enable it to be as
05:31reliable or even better than this drive
05:33yeah I don't have too much data to this
05:36but I think on the mobile phones we see
05:38that you obviously have to optimize for
05:40power consumption right like you have a
05:42battery you you want to hear optimize
05:45for battery life and so this the effect
05:51of course of that is that there is the
05:53most innovation happening there if you
05:54can use this innovation to drive your
05:56data center where you're limited to a 65
05:59megawatt now playing basically right
06:01yeah you can you can you can put a
06:03million cell phones into this data
06:05center but two million however many I
06:06need and and just get actually much more
06:09compute out of it and I think I think
06:11with flash the same right like for flash
06:13you probably don't need much less power
06:16than for spinning disk we have we have
06:19mechanics involved maybe um you know one
06:22of the one of the questions on our minds
06:24all the time I'd love to get sort of
06:25your thinking on this is how do you
06:28think the data center revolves and
06:29meaning is it going to be on pram like
06:33do people take all their existing stuff
06:35and move it over to this new
06:36architecture is it sort of private
06:39clouds and public clouds off per
06:41them what do you guys think about that
06:43and how do you you know kind of for
06:44enterprise customers which both of you
06:47are you know selling into what do you
06:49think happens in the future what does
06:51that look like so I think it's going to
06:54start out by by some companies deciding
06:57okay we are going to have our on-prem
06:59solution and we're going to have some
07:00scale out capacity in the cloud but I
07:02think going forward it'll be much more
07:04fluid it'll be I think I think as we are
07:07seeing already with Amazon there's a
07:09spot market right for for compute and
07:12for storage for entire for your entire
07:14data center essentially and I think as
07:16as we see that with more and more
07:19companies evolving we're going to be
07:23we're going to be pricing jobs
07:25individual compute jobs rather than just
07:28providing a static pool of resources I
07:30think I think that's where we're going
07:33clearly yeah I've no doubt that that
07:35over time the cloud model will take
07:39share away from the private cloud the
07:42private cloud will be there for for a
07:45very long time specially for larger
07:47enterprises because the the performance
07:50and the control is required in a lot of
07:54cases and the public cloud doesn't
07:56provide all of those I think that the
07:58most the model that will take place is
08:02probably in most cases a hybrid model
08:05where some applications some use cases
08:06are going to be on premise and many of
08:10them will be in a cloud but nevertheless
08:12the data center whether it's on premise
08:15or a data center of a cloud provider is
08:18going to follow the same theme which is
08:20communitisation of hardware the best
08:23available hardware the best in terms of
08:26price and performance and I have no
08:28doubt in my mind that because of the a
08:32you know numbers of mobile devices the
08:37components coming from the model will
08:39eventually take over in the same way
08:41that the you know 10-15 years ago the
08:44stuff coming from the PC took over the
08:47mainframe in the mini computer but in
08:49terms of whether it's going to be cloud
08:52or on-premise I think over time more and
08:55move to the cloud but I don't think the
08:57private cloud is going away anytime soon
08:59both of you are CEOs of startup
09:03companies and you know startups
09:06typically sort of exist in the cracks of
09:09where incumbents can't get to and then
09:11hopefully startups like oh I'm hoping
09:13that both you guys become really big and
09:15you become the next franchise in your in
09:18your spaces but tell me what like you
09:23know why can't why can't incumbents make
09:25it over and are you guys afraid of
09:27incumbents sort of you know coming into
09:29this new data center how do you think
09:31about that and you know are there
09:33examples maybe where they say like
09:36there's a new space here and we're gonna
09:38go dominate that or incumbents kind of
09:41you know have a good shot I'm just sort
09:43of wondering what happens to the all the
09:45existing companies in the space given
09:47this transformation into the new data
09:49center design yes so I think the way the
09:52way we see it is that the legacy that
09:55was built over time created all kind of
09:58boundaries and so for example there are
10:01several companies and then our storage
10:03companies there are networking company
10:05and all those boundaries made sense in
10:07the past and a lot of commonly build our
10:10entire business model based on those
10:11boundaries and this is what really make
10:15them thrive and I think that what we
10:18will see going forward the
10:20commoditization of hardware will create
10:22a lot of convergence where a lot of the
10:25software component will be converged to
10:27run on the same single platform and this
10:30really create a new model and many of
10:32the existing companies will have to
10:35cannibalize themselves if they want to
10:36get in which is always very difficult so
10:39this is a typical innovators dilemma
10:41they may know that they have a problem
10:44but in order to solve the problem they
10:46need to give up most of the business
10:48which most company just cannot do
10:50because of the cost structure because of
10:52cold kind of other reasons so I think
10:54there is a huge opportunity for software
10:56company the commoditization of hardware
10:58the fact that everybody realized that in
11:01the future data center will be built
11:04with commodity hardware and all of the
11:08software and converging the software to
11:10run on the same platform get rid of a
11:13lot of the legacy opens up the
11:15opportunity for startup but it's not
11:18easy for an existing company that built
11:20their entire model on those boundaries
11:22to basically compete on the same level
11:25yes I think if we look at companies like
11:28VMware for example we ever started out
11:30with providing VMs for Windows
11:32essentially and yeah
11:34we're seeing this tectonic shift right
11:36now in data centers from Windows over to
11:38Linux yeah and obviously Linux can
11:42actually deal much better with with
11:44these components like ARM processors and
11:47and we're seeing this on one and so of
11:51course VMware has to actually get into
11:52this market and capture more and more of
11:54the Linux workloads at the same time
11:57VMware is also moving or I think they
12:01will be moving across different types of
12:04applications so Sun grid engine for
12:07example doesn't generally run on on
12:09VMware I mean of course it could but
12:12HPC generally doesn't HPC workloads
12:14generally don't run on VMware and what
12:18we Emmer has done they have really build
12:20a control panel for your data center
12:21right with with a vSphere product line
12:23and they've been trying to integrate
12:27more and more areas of off your data
12:30center into their control panel I think
12:32with the acquisition of nasira their
12:34track obviously trying to get into the
12:37Sdn space so and yeah I think we're
12:43gonna we're going to see a bunch of
12:44these big companies really going into
12:46this but at the same time I think
12:47there's a lot of opportunity in in these
12:50smooth and in in the smaller space cloud
12:53I think is still pretty small actually
12:54and I think that's where that's where
12:56startups can can really execute much
12:59better I guess you're bringing a very
13:02good example so if you take a VMware as
13:03an example VMware actually build their
13:06business by taking it away from a
13:08company like Microsoft okay so and it
13:11was a great business model but even
13:13VMware as a legacy company those days if
13:16we think about Linux container they may
13:18take over because all of the MMO
13:21around the idea of virtual machine and
13:23virtualization and all the stuff around
13:24it you move to a new model which in a
13:27lot of ways is much more efficient you
13:29know how does VM are going to compete
13:31against Linux container are they going
13:33to give up their existing virtualization
13:35business in order to build a Linux
13:37container it's a very difficult decision
13:39for a company like Vemma but a new
13:41startup that has no legacy has no
13:44problem you know ditching all the old
13:46stuff and saying ok Linux container is
13:48the way to go maybe they'll just start
13:50giving away is X is X for free right
13:53yeah that'd be that'd be that'd be quite
13:56traumatic on their bottom line I'm sure
13:59speaking of Linux and containers and
14:03kind of the kind of trends there flow
14:07maybe talk about how open-source is
14:09impacting your you know the the data
14:12center and some of the innovations
14:14occurring yeah yeah so I think if we
14:17look at containers right now and docker
14:19specifically but just gaining a ton of
14:21traction the the roots there are
14:25actually well they're the roots are lie
14:28actually at Google so four or five years
14:30ago Google country contributed C groups
14:33to the to the Linux kernel and C groups
14:37are the foundation for what is now
14:39right it provides the isolation the
14:41really strong isolation in the Linux
14:43kernel without the need for spinning up
14:45a VM and docker has build really
14:48fantastic tools around C groups
14:52allowing you to really well essentially
14:55have the same benefits of a VM where you
14:57can just package your application and
14:58just run it anywhere but on top of these
15:02top of these lightweight abstractions
15:04without requiring the full VM image to
15:07be shipped and I think that's I think
15:09that this these building blocks that
15:12enable you to build really large complex
15:15systems that are that essentially can
15:18solve big problems that's really
15:19powerful and we are leveraging that as
15:22well we are leveraging C groups and n
15:25containers and docker for us this is a
15:27really pluggable model and yeah
15:31isolation is really important when
15:32you're trying to get the most out of
15:34your data so now you want to run
15:35multiple workloads on the same machine
15:37and you need something to isolate yes I
15:41mean I mean the way we when we talk to
15:44people in the data center and and yes
15:46Linux is open you know open source and
15:50it is taking a lot of the operating
15:54system market sure now we see the same
15:56earth trend in virtualization so VMware
16:00and hyper-v definitely out there but KVM
16:03which is open source in many cases
16:05especially with cloud provider is the
16:08main virtualization scheme so basically
16:12the software company that they try to
16:14build their business on proprietary
16:17infrastructure are seeing some you know
16:20competition coming from open source when
16:23the open source is good enough and
16:25pervasive enough it does take over it
16:28may not be as good for example if you
16:30look at KVM and and vmware and you look
16:33at the features and function but if it's
16:35free it has a lot more value and people
16:38are building a lot of the stolid
16:39software stack on top of open source
16:42rather than on top of proprietary so
16:45yoram that you talk that way and about
16:48open source your company is mostly a
16:52closed you know as a proprietary you
16:54build on some of these building blocks
16:55and flow your company is open source and
16:58building some proprietary so so I'm just
17:00sort of curious as to how you
17:02rationalize that thinking your am like
17:04hey if open source is gonna eat the
17:06stack why doesn't it eat you guys one
17:08and then flow when your responds how
17:11he's got his thing going how does how do
17:14you guys actually see open source being
17:16a viable business model for you if
17:18everyone uses his stuff for free so
17:20maybe let's just talk about no it's a
17:22very good question so so so I believe
17:25that open source makes a lot of sense at
17:28some level of the stick at the very
17:31basic just not a year if you look at
17:36operating system look at look at what
17:41with Linux some red that makes a lot of
17:43money so there the reason is the very
17:46basic operating SIG's stuff is something
17:49that open-source makes sense but all of
17:52the services around it people are paying
17:54money for it so at the end of the day
17:56the customization and make it work in a
17:58certain environment is something that
18:00people value so I believe that
18:03open-source makes sense where you are
18:05creating some platform that can be
18:08distributed and makes sense you know at
18:10certain level but it's some level you
18:13know the the proprietor makes sense now
18:15over time things go up so whatever used
18:18to be proprietary over time may become
18:20open-source you just need to keep up one
18:22level on top of it and providing more
18:25value that anybody else can at your
18:27level so it just doesn't make any sense
18:30for us to ignore open source it's there
18:32it's going to be there and what we view
18:34as our value is providing better
18:37customization and solution on top of
18:39open source yeah and being one step
18:42ahead of everybody else knowing that one
18:45day even our piece may be eaten by open
18:48source and we need to step one you know
18:50go one step further yeah so as Peter
18:54pointed out we are firm believers and
18:56open source most of most of what we have
18:58released today it is open source yeah
19:00and I think it really allows us to get
19:02traction it allows people to try out our
19:04software without having to sign a
19:06contract and I think where we can
19:11provide additional value that we can
19:12charge for us in making this open source
19:16easier to use and coming from from
19:19companies like Airbnb and Twitter where
19:20UX is really essential I think we can
19:25apply the same to open source and I
19:26think traditionally open source hasn't
19:27been very good about right about UX
19:30right like I mean if we think about the
19:32first distributions of Linux for example
19:34I mean the interfaces were just were
19:36just horrible yeah and nobody wanted to
19:38use it and Microsoft had this finished
19:40product and I think that's really where
19:42we can actually build some something of
19:45value for the user and I think for that
19:48you can charge even if the core is open
19:51and I think I think it's important that
19:55as we as we are seeing now with with a
19:57lot of with a lot of security breaches
20:00for example it's great if everybody can
20:02look at the at the source mm-hmm if
20:04people can discover what's which
20:07algorithm is used for encryption that's
20:10that's great it'll it'll lead to much
20:13less too much less disastrous but you
20:16don't need that you don't need this
20:18visibility necessarily in the in the UX
20:20layers and I think the UX is actually a
20:22lot of work I mean that's getting UX
20:25right is really it's an art and and and
20:28typically system companies have
20:31notoriously done a poor job in UX it's
20:34always been one of these things where
20:36like your assistant engineer and
20:38somebody else is gonna worry about the
20:40UI because I got system problems to deal
20:42with and so I think you're probably
20:44right like combining some of those
20:46elements together and thinking about the
20:48design and form and function of how some
20:50of this operates in the next generation
20:53of data center companies I think that's
20:55going to be ever ever ever more
20:56important absolutely I think when when
20:58we started we talked about the consumer
21:00at the hardware level yeah I think we
21:03need to talk about consumer at the
21:04software level I think that part of
21:06being a consumer company is providing
21:08very interactive and very efficient and
21:12easy to use user interface yeah and I
21:15think that over time the data center is
21:17going to move into the same level of
21:19sophistication at the user interface in
21:21the past it was all command line nobody
21:24cared about yeah but as the same user
21:26that are used to a very nice GUI on the
21:29mobile devices are not going to accept a
21:32user interface that is not on par so I
21:35think that on the software side
21:37delivering software the software needs
21:39to be good from a system level point of
21:41view but at the same time it has to
21:43provide the same user interface it needs
21:46to work on every device that is out
21:48there and you cannot ignore the user
21:50interface even if you're a system
21:52company even at the end of the day
21:53you're providing service to an
21:55enterprise yeah that was a great point
21:57so maybe to just wrap up here
22:00not only will mobile hardware become the
22:02basis of the next generation data center
22:05but it feels like you guys also agree
22:08the interfaces and design elements that
22:11we've all come to know and love on our
22:13mobile devices may also be a pretty
22:16critical part of what happens in the
22:18future so um flow and Yoram thanks again
22:22and we'll talk with you all soon thank