00:00hi it's Chris Dixon this is the a 16z
00:02podcast I'm here with my partner Peter
00:04Levine who so Peter you focus a lot on
00:07data center technology yeah sort of when
00:10people talk about the cloud yeah what is
00:12this cloud you someone has to build it
00:13um it seems like it's an exciting time
00:15in that area like can you can you kind
00:17of explain why yeah it it sure is and
00:20I'm Peter hello everyone it really is an
00:24exciting time you know data center which
00:27tends to be the plumbing of the internet
00:29has been one of these pretty state areas
00:33for a long time and all of a sudden
00:34there's been this unlocking of lots of
00:37innovation happening in the data center
00:39largely on the back of and not unco
00:42incidentally on the back of the whole
00:44mobile revolution like as every human on
00:48the planet starts to use a mobile device
00:50you know tweets and and emails and
00:54applications all that information has to
00:57go somewhere and it goes to back-end
00:59data centers that goes to the cloud and
01:02the cloud is comprised of data centers
01:05and computers and that whole
01:08infrastructure is now transforming to
01:10support massive numbers of users that
01:13are that are out there and eventually
01:15Internet of Things to support all the
01:18endpoint devices that are out there can
01:20you kind of explain just like so in the
01:21old days or whatever five years ago yeah
01:24if you're you would go as a company and
01:26you would you would rent space from
01:28Rackspace or so or whatever some
01:30physical data center you would install
01:32your own servers you would buy software
01:34you'd install it there like Oracle and
01:36all this kind of software right mm-hmm
01:37and now and then what happened is people
01:39started with sort of you you Salesforce
01:41and they managed their own data center
01:42because software's so software the
01:43service is the first enabler right yeah
01:45and then and then what you really had
01:48then these giant scale companies like
01:49Facebook and Google who who had such a
01:52new level of scale that no one had ever
01:53they had to actually design new
01:54infrastructure right and so can you kind
01:56of explain like how that has evolved
01:58yeah I think it's tremendously
02:01interesting what Facebook and Google and
02:03Twitter and other very large scale out
02:06companies are doing is really
02:09interesting because it's the consumer
02:12that are driving this back-end in first
02:14again they have to they have to write I
02:16mean they're supporting billions of
02:18users this whole notion of multi-tenancy
02:20you think about all these users they're
02:21all kind of you know independent it's
02:24kind of crazy to think Facebook is more
02:26sophisticated and networking right now
02:27than Cisco right I mean is that fair to
02:29say yeah it's very fair to say I think
02:32they they Facebook and Google have
02:34absolutely pushed the needle on back-end
02:37data center infrastructure and it's
02:40interesting it's all about software
02:42it's about commodity components coming
02:44being put together in these data centers
02:46so it's much less about thinking of a
02:49big vendor coming in with a refrigerator
02:51size switch or storage device it's much
02:54more about aggregating very inexpensive
02:57components at massive scale using a
03:00layer of software on top that can unlock
03:02the value so in the old days you would
03:04go to EMC for your storage and they make
03:06a custom piece of hardware and software
03:08yeah you go to Cisco for your networking
03:10yeah you'd go to what HP Dell for your
03:12servers you'd go to Oracle for your did
03:14all the systems hardware now people are
03:17saying no let's just take completely
03:18commodity and Linux and it's all
03:20software absolutely we we like to say
03:22the mobile supply chain is eating the
03:24data center right and it's it's rather
03:27interesting that as you know what mobile
03:30the innovation in mobile and the
03:32innovation and hardware on mobile
03:33devices has been phenomenal over the
03:36past several years we will be able to
03:39take many of those in Facebook and
03:41Google are starting setting an example
03:42here to take many of those components
03:45whether it's ARM chips or flash memory
03:47and take those Hardware components and
03:50aggregate those together with open
03:53source and proprietary software to
03:55create the next data center so my
03:58smartphone is connecting to the cloud
04:00and the cloud the data center is
04:01literally going to be a bunch of
04:02reconfigured smartphones I that's a
04:05great look I have this visualization
04:07that the data center is a billion
04:08smartphones connected together through
04:10very sophisticated software and it
04:12doesn't you know may not have a screen
04:14on it or whatever but it's it's ARM
04:18processors ARM processors read out that
04:20power is the most important is the
04:22scarcest resource and data center same
04:24on the answer power and cooling are
04:27critical elements in designing and a
04:29data center these days so to the extent
04:31that we can go design a data center that
04:34uses much less power requires much less
04:36cooling as optimizing on floor space
04:39those are the ingredients for massive
04:41scale out data centers and those are
04:43that those are the drivers so to the
04:45extent we can use mobile components for
04:48that it's definitely going to solve that
04:50problem so what so how does this then
04:52relate to our investing so you know I
04:55know would probably want a quarter maybe
04:58even a third of our investments or
05:00around data center technology um can you
05:03can you kind of walk through like
05:04different areas of the storage
05:05networking computing yeah so on a on a
05:09on a broad level if we take Facebook and
05:11Google those companies are becoming the
05:14blueprint for data centers of the future
05:17it used to be where companies would look
05:19at Wall Street for example at fidelity
05:21and Morgan Stanley and say hey we want
05:23to copy we're gonna build a data center
05:25like that melody would build something
05:27and maybe some would be custom and then
05:29and then the companies would come along
05:31and say you know let's take that
05:32factored it out and sell it to other
05:33people and and right companies would
05:36come exactly so it was a startup startup
05:39would go look at fidelity or or Morgan
05:42Stanley and go build a stack of
05:44dangerous an entrepreneur looking to do
05:46data center company you'd go and Lerma
05:48fidelity is doing package it up as
05:51merchants market as merchants market
05:54products to make it available to the
05:55rest of today you would go look at what
05:56Facebook and now you go look at what
05:58Facebook's doing and Google's doing and
06:00you say ok instead of me as a company
06:03having to need 10,000 engineers to keep
06:06this up and going how can how can we
06:08create the merchants market products to
06:10make that type of data center available
06:12to data centers around the world to
06:15companies around the world who want to
06:16go build this next generation of system
06:20infrastructure and so a lot of the
06:22companies that we invest in whether it's
06:24storage networking compute databases
06:28analytics security all follow the model
06:32of how do we unlock the value of
06:35commodity hardware components with
06:36sophisticated software to
06:38this next generation scale-out Data
06:40Center to everyone who's building that
06:43so we can talk about a few of these
06:44companies so like on a networking level
06:46nice era it was a company that we yeah
06:48that's didn't can you explain what they
06:50do yes a nice era a company that was
06:53acquired by VMware they really pioneered
06:57this whole notion of software-defined
06:59networking and really unlocking this
07:02sort of hard-coding of of networking to
07:07something that's much more flexible
07:08anytime you build something in software
07:09it becomes inherently more flexible so
07:12they unlock that and what's the minute
07:15what's the customer benefit the customer
07:16benefit of that is that no longer am i
07:20locked into particular silos of
07:22networking or having to lock the honest
07:26a a server to a particular underlying
07:29piece of storage or a user to a
07:31particular server it's very fun jables
07:33and so if I'm if I'm creating a
07:36service-oriented architecture I need to
07:40be very flexible in terms of how my
07:42network connects to different parts of
07:44my physical infrastructure and that the
07:46software layer actually enables us to go
07:48do that we so nice era is now part of
07:51VMware and VMware is taking that that to
07:55market we are on the networking side one
07:57of our current investments as cumulus
07:59networks which is building a very much
08:02along the theme of replacing network
08:05switches what they are doing is taking
08:08commodity Hardware putting a layer of
08:11open source and closed source software
08:13on top of commodity servers and turning
08:17that into a new switch that is as fast
08:21as is you know as capable as some of the
08:24most proprietary systems that are out
08:27there today from the likes of Cisco and
08:29juniper hmm and then what about storage
08:32so on the storage side we see much of
08:34the same thing where again it's all
08:38about software being built on top of
08:40commodity components whether it's flash
08:43or you know flash has become a huge
08:46trend in storage while unlocking that
08:49flash in using commodity flash as the
08:52basis of an enterprise storage array is
08:55quite compelling for the for the market
08:59to the extent that we can build you know
09:02take software and unlock the value of
09:04commodity flash we can literally see in
09:07the not-too-distant future
09:08the cost of flash storage devices being
09:13in the same price as you know spinning
09:16media and that's going to be very
09:17compelling it's fast it's you know
09:19reliable so I think that the
09:23transformations on that side are really
09:26interesting as we start to use again
09:28these commodity components for the basis
09:30why weren't people using commodity
09:31components let's say five or ten years
09:33ago it wasn't the performance wasn't
09:35there Moore's laws things so so one
09:39could argue that we did use commodity
09:42components you had mainframes mainframes
09:44were replaced by PCs so five ten years
09:47ago it was the PC revolution so x86
09:51chips and you know disk drives from a PC
09:55and Microsoft software and Oracle did
09:58that was that was quote-unquote
10:00commodity commodity now is even way more
10:03commodity in that in the same way that
10:05pcs replaced mainframes
10:07I see mobile mobile devices replacing
10:10PCs so it's really you know in ten years
10:15ago people were talking about commodity
10:16components in the data center meaning
10:18x86 and PCs well now we'll take that a
10:21step further power cooling the price of
10:24these components coming down that much
10:26further it's just the further
10:27commoditization of enterprise Hardware
10:30that that that is continuing and so what
10:33and and and so you know how you think
10:36the world can look in five to ten years
10:37in the sense of is every large company
10:40going to be building their own
10:41datacenter using this software are they
10:43gonna be outsourcing it to AWS and
10:46Amazon and Google like the stuff that
10:47we're seeing now is start like every
10:49startup we meet right yeah it would be
10:50strange to not be running your stuff on
10:53yeah yeah look I think a lot moves to
11:00the cloud I think there's the big debate
11:01of does everything move to one but
11:03there's sort of private or public cloud
11:05there's private you know there are
11:06clouds Goldman Sachs has their own quote
11:09cloud right right so I think certainly
11:11the notion of private and public clouds
11:15will be the dominant compute sort of
11:17style in the future so there won't be a
11:20question as to whether you do a cloud
11:23quote-unquote cloud style infrastructure
11:25we're talking about it's so much better
11:27but it'll be a question as to whether
11:28you own it and control it correct
11:30correct and that will be around secure
11:32security data you know Kayapo and all
11:37all of you know and and and so and you
11:40know sort of DNA's of the company
11:42themselves on how they view this but
11:44every data center their future will be a
11:46scale out data center right applications
11:48are gonna change the user base is going
11:51to need that as you know again mobile
11:53and then we start to think about
11:54Internet of Things as you know
11:56collecting all this information and
11:57analytics that needs to be done on that
11:59all point to this notion of the scale
12:02out data center whether it's public or
12:04or private so you think it'll be and the
12:07difference will just be as you said
12:08depending on the the DNA of the company
12:11or maybe some security issues or
12:13something but there's not a whole lot of
12:14other there won't be other reasons why
12:16you would go private or mean on-prem on
12:19prim private you know this is a it's a
12:24it's certainly a you know if you have a
12:27private data center right now on prem I
12:30think it'll take a while so a lot of
12:31people will say well of course I'm going
12:33to continue to do that I think that the
12:35economics and the way that data centers
12:37are going to be built is going to drive
12:39everyone to private or public clouds and
12:41the notion of the on-prem data center
12:44starts to disappear but you know I will
12:47see what happened what and you mentioned
12:49Facebook and Google leading the way
12:50Amazon it's not leading the way I mean
12:52that's the P most people think of cloud
12:54they think of Amazon I think I think
12:57that the Google and Facebook
12:58infrastructure is now in my opinion more
13:02advanced than what Amazon is Amazon was
13:05created several many sort of already
13:10generations ago relative to this
13:12commoditization can you give some
13:14examples of the kinds of things that you
13:17you know I'm virtual machines for you I
13:20think yeah that's a good one
13:24you know Amazon Amazon from what I know
13:27is based on a virtual machine
13:29architecture and virtual machines are
13:33were you know the way to encapsulate
13:36information and encapsulate sort of you
13:41know kind of data I see that we've
13:45already moved to the next generation
13:47where you know virtual machines
13:49certainly which is containers which
13:52certainly in a Google or Facebook
13:53environment it's more about containers
13:55and less about virtualization now why is
13:57that well containers use far less
14:00compute resources they're you know it
14:02used to be where I had to support
14:04Windows and Linux on a machine therefore
14:06virtual machines are important now it's
14:09only Linux on both sides the Linux as
14:11the as the base operating system and
14:13running within these containers so we
14:15can simplify the environment by as an
14:18example through containers and that
14:20unlocks a lot of a lot of additional
14:22power for compute for applications as
14:25opposed to locking that power down for
14:27for the virtual machine infrastructure
14:30so what are the areas in the data center
14:32that we're like that we're looking at
14:33for you know new investments so what do
14:35you consider there well one of the new
14:37investments that we recently made was a
14:39company is a company Mesa Sphere and I'm
14:42particularly excited about that
14:44particular architecture in that it
14:46literally is what I call the one over
14:50vmware technology company and when i say
14:53one over vmware this is not about
14:56carving up machines and carving up
14:58resources it's about aggregating
15:00resources together and i think that that
15:03the notion of aggregation of processes
15:07compute resources storage and bringing
15:10that all together against this this
15:12massive scale out kind of data center is
15:15going to be the way that that that
15:17environments run in the future so i
15:21think of it as tell me this is wrong but
15:23i think of it as if so in the cloud
15:25architecture the data center is is
15:27becomes a lot like a server used to be
15:30and therefore a single server so yeah
15:33now the entire data center is voice and
15:36since one server and all the things used
15:38to do at the server level in the
15:39operating system are now need to be done
15:41to the data center level and I think
15:42Mesa is sort of like maybe the
15:43scheduling but it is a platform for for
15:49the aggregation of these resources
15:51beyond just scheduling so one could
15:53imagine applications actually being
15:55written to the maysa API and it will
15:59coordinate the right resources from the
16:02right places coming together not unlike
16:05you know an operating system does not an
16:07individual server you can say I want you
16:10know to increase the capacity of web
16:12servers 50% and database is 10% and
16:14it'll just mace us we'll go out and get
16:17that for you and make it available to
16:18the application just without without the
16:21user doing in the old days you had to
16:22have like a team of sis admin's you know
16:24whatever spinning up servers and
16:26configuring and your analogy of like
16:28single server to data center being the
16:30server is exactly the analogy within a
16:33given server I could always you know I
16:35could tune things up and around soon but
16:38at the data center level very very
16:40difficult to do and then things change
16:42and it's you know sort of a shitshow a
16:43stuff that goes on and it's crazy right
16:46and so as you expand out to thousands or
16:48you know we talk about a billion cell
16:50phones imagine trying to manage all that
16:52without something like a meso scalping
16:55to coordinate it all so what other areas
16:57you think like are you know that we
16:58haven't made investments in yet that you
17:00think are interesting um I'm interested
17:02in I think the whole container world is
17:06is particularly interesting
17:09so containers oh and let me do my
17:10laypersons yeah take on containers too
17:12so you know virtual machines if in the
17:14data center world had probably been one
17:16of the biggest inventions of the last 10
17:19it's that VMware massively successful
17:21company pioneered it it basically allows
17:23you to run an operating system on top of
17:25an operating system which is great for
17:26both running Windows on top of Linux but
17:30also in in a way to partition resources
17:32effectively so you consider say I write
17:35I mean it lets you it lets you change
17:36things in software they didn't pass you
17:38to change in hardware like you want to
17:40run more web servers or something and
17:41but the problem was you're running an
17:44operate system which is a preposterous
17:45architecture it's a fun well it's a
17:50complete replica like that operating
17:52system the base level optimization suite
17:55it's a complete reproduction of the
17:58server itself it's like and that
18:00everything is like dream within a dream
18:04you know resource intensive it's very
18:07resource intensive now so so containers
18:09are you get the benefits of that but you
18:11you're not doing a dream within a dream
18:13you're doing one layer one dream right
18:16yeah that's one it's sort of one layer
18:18at the operating system effectively
18:21reproduces itself right so you have
18:23multiple instances of the same operating
18:26system running on a system and look I
18:29think the notion of of canoe the notion
18:34of having a container and being able to
18:36move things and this was the the best of
18:38the virtual machine world was that I had
18:40this this entity that I could move
18:43around and it was like so a developer
18:46now what happened also right is in the
18:47old days you brought a program you're
18:49out a program right now you write a
18:50program you got 50 different libraries
18:52you've got a database you've got a web
18:54framework and just getting the thing
18:56deployed like so you take your software
18:58and I want to put it on a new machine is
18:59like a day of work of like downloading
19:02and figuring and stalling yeah and so
19:04containers let you package it all up in
19:05one single file and boom you can see my
19:08shoot it anyway so and that's been that
19:10was that is the value is amazing
19:12so for virtual machines that was this
19:14you know had similar characteristics one
19:16file I could go deploy in multiple
19:18places containers offer that without all
19:21the overhead of this virtualized
19:24infrastructure and so I think that's
19:26what it is it's a rare thing where it's
19:28it seems like it's a no-brainer dramatic
19:31improvement over virtual machines and
19:32their dreams are over it's gonna take a
19:35while don't take it certainly will take
19:37a while but there's just there's not
19:39there's not really trade-offs here the
19:41well the trade-off in the past was and I
19:43mentioned this before is that what
19:45virtual machines allow you to do is run
19:47different operating systems on top of
19:49the base operator that was a big it
19:51doesn't matter anymore so while it's a
19:53no you know Linux well I know people
19:55they don't understand us it's so
19:56thoroughly it's pervasive right it's one
19:59and so if I happen to be running Linux
20:02on top of Linux like hey that's just
20:04fine right in the in the quote-unquote
20:06old days I to run versions of Windows I
20:09had to run different instances of Linux
20:11and so that mixed environment created
20:13the the the the opportunity and
20:16rationale for for server virtualization
20:19so I looked I think the world is is
20:22shifting in this new direction and you
20:25know it's gonna unlock I think a lot of
20:27value out of you know these data center
20:29infrastructure components so around
20:31containers so we have Mesa sphere like
20:32what are the other is are interesting
20:34kind of startup opportunities are around
20:38Michener's as being so so one of the
20:41things that I'm particularly interested
20:44in is um is the hosting of a variety of
20:52infrastructure components so we made an
20:54investment in a company digitalocean
20:56right and what they do is they take a
20:58variety of open-source components and
21:00they you know they're they're a cloud
21:02for developers and what I find really
21:05interesting about their environment is
21:09they are offering to developers the
21:11hosted instance of many open-source
21:14projects so I'm a big believer that
21:18we're moving from a world of sort of you
21:22know this on-prem kind of you know
21:25virtualization running on-prem to these
21:27hosts and instances of let's say host an
21:29instance of docker for example or hosted
21:32instances of databases and so I think in
21:36my mind to the extent that we look to
21:39invest in the in in the not-too-distant
21:41future it'll be about investing in
21:44hosted instances of particular open
21:47source technologies so our hosted docker
21:49might be an interesting thing this is
21:51partly because it's it provides a yes
21:57one it provide one it provides great
21:59value to the end customer because I no
22:01longer have to deploy this stuff on Prem
22:03right and then the business of open
22:06source is completely
22:08nationalizes open-source as a viable
22:10business model in the past open source
22:12was a questionable business model
22:14yes Red Hat made a you know nice way out
22:17and I wrote a blog on this like you know
22:19there are some companies who will make
22:21good money with successful but there
22:24haven't been other there haven't been
22:25different acquisitions but there haven't
22:27Minister haven't been standalone
22:29franchises other than Red Hat that have
22:32really been able to quote-unquote
22:33you're gonna type in this scales or the
22:35importance of open source given that
22:37it's as if you know if you compare us to
22:39its to its proprietary software
22:42analogues it's you know you just you
22:45just made a couple of minutes ago you
22:47made the point that Linux has taken over
22:50the world right there's no need for
22:51anything else well if you think about
22:53the economic benefit that Red Hat has
22:56derived over the complete transformation
22:58of Linux it pales in comparison to the
23:01impact that Linux has had in the world
23:03and so what open source has been
23:06fantastic it's been a fantastic vehicle
23:10for innovation but in terms of economic
23:14value back to the developer of open
23:16source and back to the provider of open
23:18source I believe that once it once open
23:21source can be provided as a service it
23:24will have it is then able to unlock full
23:28market value and full potential and to
23:30me that's a really interesting sort of
23:32transformation so does that mean the
23:34company like historically with open
23:36source the interesting company has been
23:38the company that had the core developers
23:40of the project so for example my sequel
23:42the company had the my sequel developers
23:44right like yeah who's that and so in the
23:47future you think that there's there's a
23:49cool new open source project ax the key
23:52developers for X will be in a company
23:54and a company will provide a service
23:56based version yeah of that software
23:58that's what you think will be the future
24:00yes yes I think that's that to me is the
24:03interesting part of the next gen open
24:06open source models will be that
24:09particular that particular type of
24:11deployment and you know github is a good
24:14example of this like you know they're
24:16the they're the epicenter of all open
24:18source development of all development
24:21and they can get and you could have sold
24:23it in my sequel style or you could do
24:24github which seems to be a much duh
24:28right like yes so that model is exactly
24:32what I believe is going to happen at
24:34every segment of the open-source market
24:37will be a we get to get have the you
24:39know github has that network effects yes
24:41do the other I mean that that interfere
24:44if it's really the beginning of a
24:46pattern or if it's an accession
24:47exception I think look github certainly
24:51has the the the the network effect of
24:54developers but I would argue that if if
24:57they had just gone often and supported
24:59get and done it as an on-prem thing with
25:02some collaborative who on top of it it
25:04would never have been as successful as
25:06what it is now so sure the collaboration
25:09adds to it but look I mean I think you
25:12know there are a number of companies who
25:16I believe will have very great promise
25:18of offering sort of the sassa fide
25:22versions of open-source that will be the
25:24future just because it's going to be
25:27very compelling for them to continue to
25:29innovate and for users to actually use
25:31that in ways that we just haven't seen
25:34all right great we're out of time thanks