00:00hello this is Peter Levine from
00:01andreessen horowitz and I'm here with
00:03Ramona and junior Ramona is the CEO of
00:07coho data and JR is the CEO of cumulus
00:12welcome guys what I'd what I'd love to
00:16chat about is is the future of the data
00:19center and specifically the
00:22consumerization of enterprise hardware
00:26and what that means for the data center
00:28and how you guys think about that so
00:30I've gone out and I've talked about sort
00:33of this you know this notion of the
00:35mobile supply chain eating the data
00:37center and you know I have this big
00:40belief that mobile the components of
00:43mobile phones will become the basis of
00:45the next generation data center really
00:48really inexpensive sort of you know
00:51cheap components that require a lot less
00:53power floor space cooling but all of
00:57that stuff being sort of connected and
01:00propagated together with sophisticated
01:03software and so I'd love to maybe get
01:08your guy you know your guys opinion on
01:10this and maybe describe what's happening
01:12now in the data center you know kind of
01:15is it an incremental change is it a
01:17wholesale change what are you seeing out
01:20there and how are you seeing some of the
01:22trends rahmatu maybe start with you and
01:24then junior you can absolutely thanks
01:26Peter great to be here definitely there
01:28is a lot of consumerization I'd like to
01:31call it sort of the breaking down of the
01:33data center I come from a storage
01:36perspective right and the past 30 years
01:39storage purchases have roughly been the
01:41same which is it's like buying a car
01:42it's a lifestyle choice
01:44you know your performance then you go
01:46buy a high-end storage device or a
01:48low-end storage device you silo yourself
01:51and then you complain about utilization
01:52I think that's all breaking apart what's
01:55happening is instead of people trying to
01:57buy for like three to five years in
01:58advance they are thinking about let me
02:01buy what I need right now you can't do
02:03that in the traditional model that
02:05that's where first it's breaking down
02:06the refrigerators full of discs breaking
02:09down into smaller components now people
02:12are saying hey you know
02:13have visibility into my business for the
02:15next year so I'll buy like storage for
02:18the next year it's up to you to
02:19incumbent on you as a vendor to make it
02:22easier for me to scale from there add
02:24another box add another box this is
02:26where the value is not as much in
02:29Hardware customizations anymore it's
02:33about commodity components and letting
02:35the software be you know and letting the
02:37software be care capable of
02:39orchestrating ease of management and
02:41ease of scale it's incumbent on software
02:43and the other thing which is catalyzing
02:46this is open source and you know you
02:49have been you have commoditization of
02:51hardware and you know you're talking
02:52about the consumerization you have a lot
02:55of devices you have a lot of endpoints
02:57and you also have a lot of you know like
02:59on the server side buses and disks and
03:02flash and so forth and it's almost
03:03impossible for anybody to have for a
03:06single entity to develop device drivers
03:09and stacks and everything for that
03:10it's actually a spread out development
03:12model and open-source is key to that
03:14yeah every company is contributing to
03:16open source and keeping a platform a
03:19software platform on which you can build
03:21more value adds they're contributing to
03:24it as a result of which innovation is
03:26just taken over kind of building on what
03:29Manas was saying is I don't know that a
03:32hundred percent agree with you that the
03:34components in the cell phone or the
03:36mobile the mobile you know devices are
03:39going to drive the data center the
03:40future but I think a lot of the tenants
03:41that allow those to exist hold true so
03:44and you're like give me some example
03:46yeah exactly so I mean some of the
03:47really key pieces are Ramana said
03:49looking forward and buying from my
03:51business over the next three to five
03:52years what we understand is you know if
03:55you look at cell phone technologies a
03:56lot of them kind of modern technologies
03:58around ite there are a lot of industry
04:01standard or well-known industry
04:04mechanisms in x86 processor or an ARM
04:06processor or flash memory it's disks
04:09these things are well understood so
04:10people can put these together and again
04:12in well understood ways kind of again to
04:14Ramallah Romana's point around
04:15open-source it's it's kind of the whole
04:17industry at large moving at different
04:19rates with each other putting things
04:20together at this moment in time
04:23recognizing that two years from now it's
04:26going to come together
04:27little bit differently so this allows
04:29customers you know both you know
04:32web-scale type you know or web facing
04:34businesses as well as enterprises to
04:36acquire and consume that technology on
04:38an as-needed basis and recognizing that
04:41i can get you know effectively what i
04:43it all comes together what I need in
04:46five years is going to be very different
04:47yeah yeah that's that's interesting and
04:50when I talk about just for clarity here
04:52like ARM chips and flash to me are
04:55innovations that really started in the
04:59mobile world and are moving over to the
05:02enterprise world it's not unlike you
05:05know we saw many years ago the shift
05:08from mainframe to PC right mainframes
05:10where in the data center pcs were on the
05:13desk at someone's house well the PC
05:16innovations and revolution eventually
05:18took over the data center and I think
05:20that it's the innovation is now that
05:21you're physically gonna have a mobile
05:23phone data center right you'll have
05:25components that have been influenced by
05:28the development of mobile phones that I
05:30think really it's already transformed I
05:32mean storage as an example using flash
05:34like you don't have a spinning disk in
05:36your cell phone right but let's I think
05:38what's kind of slightly counter to that
05:41or not encounter it's it's a line but
05:43not 100% directly aligned is the fact
05:45that I agree with you around the fact
05:47that the the proprietary systems in
05:49existence at a Stennis Center will open
05:51up much like the mainframe PC PC took
05:53off because it was a grossly open
05:55architecture everybody understood
05:57generally how it worked and people could
05:59innovate it on an on it at scale
06:01yeah whereas if I was IBM and I did the
06:03mainframe guess who got to innovate on
06:05the honey IBM IBM right yeah well and
06:08not only did a lot customers in which is
06:10bad for the customer but it also locked
06:11up innovation didn't allow the industry
06:13to move an industry pace you know one of
06:16the most interesting things that that I
06:18find now is you know IBM db2 which was
06:21the reason you bought an IBM mainframe
06:23now runs on every processor architecture
06:26across every operating system and it
06:28performs better on Linux x86 than it
06:32does on the IBM mainframe today in terms
06:34of transactions per second yeah and the
06:37other thing is well raid which is when
06:39you talk about consumerization
06:40people say hey is I can't trust my cell
06:44phone to be my next server because it's
06:45gonna fail exactly but now set about pcs
06:48exactly that they're gonna fail but the
06:51the beauty of innovation is that you
06:53will innovate accounting for failure
06:55right I mean so if you're gonna build a
06:58you know a big system you break it down
07:01into 20 components now you will just
07:03assume that six of them are going to
07:04fail and you'll try to figure out how
07:05best to recover from yeah but this is a
07:07sea change from you all right here
07:09because like the fundamental problem
07:10with that thesis is you can't drop the
07:12mainframe in the toilet yeah like your
07:16cell phone the reason your cell phone
07:17feels your cell phone is actually a
07:19pretty reliable device it's got very few
07:20moving parts and almost everything is
07:23integrated a cell phone operating in a
07:25reasonably static stable environment I
07:27bet you find the MTBF there is better
07:30than a mainframe the reason your cell
07:31phone fails is because you throw it at
07:33your kid or you know you drop in the
07:35water or something else like that
07:37yeah they I mean from my point of view
07:39ready I mean if you see flash flash
07:41actually fails more than a disk disk has
07:44been around for 50 years flash actually
07:46fails quite a lot no when you design a
07:48system right these days you think about
07:50how do you just you just account for it
07:52and you account for it in software and
07:54you become more innovative right I mean
07:56in terms of how to make sure that I can
07:58very quickly build something that is
07:59super aligned by using scale to your
08:01advantage rather than making this it
08:04goes back in our title a couple verses
08:05loosely coupled loosely coupled it's
08:07just better because you can then you
08:09don't get sort of obsessed with a single
08:11component failure so it sounds like we
08:15all believe that open components both
08:18hardware and software will change kind
08:21of are changing the data center more
08:24openness allow enables for more
08:26innovation what do you think this means
08:28for incumbent vendors and like you know
08:32okay it's obvious hey we're you guys are
08:35telling the world here that um you know
08:37it's open components and open software
08:40so what does this mean for incumbents
08:42and like what you know what's gonna
08:45happen in your mind I think the
08:46landscape changes the way I think about
08:50it which may not necessarily be the way
08:52others do is is kind of will calm
08:55people that are able you know like what
08:57we do at cumulus right we're a
08:58networking company we have a recently
09:00thin you know op super operationally
09:02efficient model where we leverage open
09:03source and provide you know great
09:05networking solutions to our customers
09:07some of our competitors which are kind
09:10of very large incumbents invest
09:12tremendous amounts of money trying to
09:13protect their value add and we kind of
09:17go out about the whole problem the other
09:18way around we strip away the only thing
09:21and leave only the things that matter to
09:23customers and get rid of the kind of
09:24excess value add and then that leaves a
09:27space for you know the supply chain
09:30meaning that you know resellers system
09:32integrators and those people to kind of
09:33come along and provide value out of
09:35services that really make all of this
09:37new technology consumable by customers
09:40and you know you might ask yourself you
09:42know what's different about that as
09:43opposed to going to a great big
09:45incumbent to build yeah by an existing
09:46system the difference is it keeps
09:48everything very loosely coupled we
09:50innovate at our pace all of the other
09:52people that come together to make a
09:53solution all innovate at their own pace
09:54is and the solution providers that that
09:57integrate it and deliver to customers
09:59are able to kind of have this continuum
10:02of offerings that make sense
10:03again to the customers creating
10:05competition and effectively you know
10:07focusing all of us on the necessary
10:09elements and avoiding all the Croft
10:11mm-hmm-hmm Raman and maybe you know you
10:15can certainly comment on that but how is
10:18this new trend different from what's
10:20happened in the past I mean I've look
10:22how we've all been around I've known you
10:23guys for a long time you know we've all
10:26been around hey open is good and like
10:28this whole notion of open has been
10:29around for a long time it promises great
10:33innovation why is this it feels like
10:35there's a real inflection point in data
10:37center design and sort of what's
10:40happening now into the future relative
10:42to sort of what's happened in the past
10:45what do you think the difference is now
10:47yeah so I called as a you know if you
10:50looked at like stacks of applications
10:52this is commoditization sort of moving
10:54up Ward open source does is that
10:57actually takes a lot of your sort of the
11:00lower end layers the things that work
11:02with hardware and just just makes it
11:04commoditized as an open platform for
11:06everybody it's just available
11:08right asked available for everybody who
11:10wants innovate and it's in the interest
11:12of hardware vendors to contribute and
11:14keep that platform robust and growing
11:16yeah that wasn't the case ten years ago
11:18ten years ago proprietary systems still
11:21held some value there was this belief
11:23going back to the arts point which is
11:26what has happened is db2 is just better
11:28on a Linux platform on an x86 hardware
11:31and that sort of a commoditization of
11:33the core you know platform is is what is
11:37driving and innovation goes to like the
11:38higher and higher levels which is making
11:40this this possible it also brings costs
11:42down to which is the overall cost of
11:46putting the solution together once it
11:48moves from proprietary to open-source
11:49not only is it easier to innovate it
11:51also brings in like cost pressures and
11:55what was you know traditionally
11:56proprietary and makes it open and
11:58available to a lot of folks and for
12:00example it was you know to think of some
12:03things like OpenStack getting the level
12:06of you know OpenStack OpenStack summit
12:09this year had four thousand attendees
12:11five thousand more than doubling from
12:13the past year it's it's going to become
12:16the next VMware to tell you the truth
12:18because that is going to be the platform
12:20for you know cloud innovation or an
12:22application innovation and so on and so
12:24forth the reason is again nobody's tied
12:26in to anything proprietary there so you
12:29know that that's the direction that's
12:31the direction these things are going
12:32anyways that combined with hardware
12:34platforms just makes it you know that's
12:35that's the data so innovation in
12:37radically yeah and I think there's the
12:40inflection point is less dramatic than
12:42you think only because you're looking at
12:44it from the outside not the inside you
12:46know having me up kind of help make the
12:48sausage some of the incumbent networking
12:51providers you know a long long time ago
12:54people would you know you built the
12:56operating system you built the all the a
12:58six you manufactured it downstairs you
13:00did everything and then pretty soon you
13:01started out seeing the the manufacturing
13:03you started using industry standard
13:06components right I don't call them in
13:08commodities they're not just like always
13:09commodity meaning there's 65 suppliers
13:11but the column industry standard
13:12components I love you yeah yeah hit come
13:15on industry standard right well
13:20understood and broadly available to all
13:22comers right that's kind of the the
13:23moral equivalent of that but the other
13:26thing that happened during it's been
13:27occurring over a long period of time is
13:29these encumber providers are using a ton
13:31of open source you know if you look at a
13:33cisco for instance a cisco operating
13:35system in x OS is there their data
13:37center class operating system a
13:38phenomenally high percentage of that
13:40code is open source they just don't tell
13:45you that's that it is Linux but it's
13:47like some super old version of Linux
13:48they don't they don't have good hygiene
13:52around what it means to be a good
13:53consumer of open source so like it's
13:55really hard to get access to their
13:57patches you know it's really super old
13:59so they they keep it a secret because
14:01they want to say hey we're the ones that
14:03provide the value-add we control the
14:04customer we can control everything and
14:06and that's the part that's kind of
14:08inflecting is wrecked as customers are
14:10starting to recognize it hey wait a
14:12nx-os is linux and i'm not getting the
14:15benefits of that because you know
14:17heartbleed is a great example heartbleed
14:19came out cumulus linux we do a network
14:22operating system focused on data center
14:24we had all the patches and all the new
14:26muse by Nerys up in about three hours
14:29from the time we found out of our
14:31heartbleed cisco took about three weeks
14:33to get it up for an EXO s because they
14:36had to back border to their antiquated
14:37version of every open source package
14:41that they were using and they weren't
14:42you know telling people about so they're
14:45kind of in this weird weird spot where
14:46they're using this stuff they don't want
14:48to tell anybody because it kinda it kind
14:50of takes all the the gold off of the the
14:52outside of the whole thing but then
14:54their customers suffer because that
14:56they're trying to maintain that customer
14:58mind share and and guess who are GRT our
15:00point right which is the next cisco vm
15:03sorry boomer that's open so soon i mean
15:06literally if you looked at how much of
15:08is based on open source it's just that
15:10you know they probably don't come out
15:12and claim it right off the bat right but
15:14it's very I mean we come from the Zen
15:16world that's our our DNA so we know
15:19exactly how much of ESX is actually open
15:21source which is good and you know good
15:23and that's the direction that's an
15:25interesting observation about using open
15:28source and sort of hiding behind it as
15:30opposed to being transparent about it
15:32and the downstream sort of benefits of
15:36being transparent to customers as
15:38opposed to hiding behind it because
15:39customers now this heartbleed example is
15:42an excellent example of how you know hey
15:45like we can fix this in a in in a day
15:47because like we're very transparent that
15:50it is open-source you can use what's out
15:51and innovation therefore increases and
15:54customers kind of get you know cycle
15:57times get reduced relative to new
15:59features and adoption curves and all of
16:01that yes that that kind of parallel
16:04undirected innovation which is kind of
16:07general thesis around open source or
16:09effectively open elements when you can
16:10for animal things together to create
16:12solutions it allows you to you know each
16:16piece to innovate it at scale people
16:18have the great ideas you know the the
16:20coolest thing for me you know as a CEO
16:22is when a customer gets a hold of our
16:25software puts it on some hardware and
16:26they they go through and say well how
16:28would you operationalize this and we
16:29give them this is our set of recipes and
16:32then you know they call back and say hey
16:34guess what your recipes were okay but
16:36look what we did yeah and you look at
16:38you're going holy God this is really
16:40really cool so this customer innovated
16:41took it a whole different direction we
16:43never would have thought and you know
16:46that's kind of what makes the whole
16:48world go around now yeah interesting
16:50maybe I'm you know let's shift to some
16:53of the business challenges and
16:55opportunities that you guys are seeing
16:58relative to data center and I'll
17:01characterize that you know data in a
17:03data center world it's often as a young
17:06company it's hard to crack into the data
17:08center stack right it's got you know one
17:11incumbents stacked on top of another you
17:14know all the way up through data bases
17:16and above and as a as a young company to
17:18get inside of that very controlled stack
17:22is a is a non-trivial problem so I'm
17:25just sort of curious like you guys are
17:28running phenomenal companies right now
17:29and you intend to become part of the
17:32next sort of data center of the future
17:35how do you see is it a different now do
17:37you still have to fit into the existing
17:39stack or is it is there something else
17:41going on okay I'll go first
17:44kort been you know ben horowitz this
17:47blogger and a price sales is still the
17:49same as it was ten years ago so you
17:52bring the new technologies and you bring
17:54everything but it's still you goat you
17:56know the best chance you have is new
17:58apps new developments in greenfield
18:01environments especially for us because
18:03we come from the storage sort of things
18:04which is probably the most conservative
18:06it's people's data right I mean they
18:08don't they don't move things around that
18:10easily but having having said that I
18:13think there's this notion of you know
18:17like cloud inspired architecture so
18:20everybody's cloud inspired now which is
18:22which is fundamentally changing the data
18:23center there are three elements I mean
18:25the wave and I describe cloud
18:27inspiration right there are three things
18:28to it which is scale on demand don't
18:31sell me what you are telling me I'll
18:33need for five years yeah
18:34sell me what I need for the next year
18:36right and pay as you grow kind of goes
18:39into that we just make it super simple
18:40for me to add one more component twelve
18:43months from now when my business is
18:44growing and it's up to you to make sure
18:46that everything is just work seamlessly
18:47so I want to be able to scale on demand
18:49the second thing is to be able to
18:51perform on demand this is a big deal in
18:54storage which is like as you added
18:55capacity performance tapered off that's
18:57the income that that's how things are
18:59that's how things have been for a long
19:01time is to perform on demand and the
19:03third thing is multi-tenancy which is
19:06you know in the Mineola days again going
19:08back to my car analogy which is your
19:10important abs you bought expensive
19:12storage connected those two for home
19:14directories you bought cheap and
19:15cheerful storage and you created these
19:17silos the whole notion of cloud breaks
19:21it's like I want to run both my IAP
19:23hungria as well as my lazier on the same
19:27storage it's up to you to figure out
19:29very each of these apps goes whether it
19:31gets flash or this these are so
19:33fundamentally right people are thinking
19:35that way anyway so for for newbies like
19:38us who are going into the enterprise we
19:41are able to in you know that message
19:42sticks around a lot of the CIOs in the
19:46enterprise are you know they use Google
19:47Mail their Facebook and all of that and
19:49they do know that that infrastructure is
19:54on you know the VAE data center so bill
19:56didn't know that but still we you know
19:59we do go into new environments new apps
20:02that's where we get the most traction
20:03yeah and and I believe or a period of
20:06time it's not gonna be right um you know
20:08next tomorrow but over a period of time
20:10you would see this new architecture
20:11tickle yeah gerrae what are you seeing
20:14and sort of on the networking side we're
20:17I mean we don't have people's data so if
20:19someone loses a packet most um pretty
20:21much every protocol these days will
20:23recover from it so we're a slightly
20:24better off place but it's it's kind of
20:26weird because the the behavior I noticed
20:29not only around Network commit again in
20:30storage as well as customers have dealt
20:34with the existing incumbents and the
20:36existing incumbents were not perfect not
20:38like they woke up in the morning they
20:39delivered perfect product and if never
20:40tripped and fell it's actually quite the
20:42contrary the customers have had lots and
20:45lots of problems with the existing
20:46incumbents and said through that set of
20:47problems and resolution they've not only
20:50come to trust the incumbent which is
20:52fair but they've also become to rely and
20:54and fear not having the incumbent like
20:57oh my god what happens if company X goes
20:59away you know something bad might happen
21:01right and there's people lurking under
21:06so what we're seeing now is enterprises
21:09you know and actually even even though
21:11also all the mega scales to you know
21:13actually all the Megane scales that I
21:15know of right now have made the decision
21:16the existing incumbents don't provide
21:17very much value so they're all
21:19transitioning away one way or the other
21:20kind of contrary with Cisco says in
21:22their X number of big data centers in
21:24the world you cisco gear you know bs
21:27let's go let's stand up the customers
21:30and talk about it right the and so
21:34what's what's happening now is the
21:35enterprise's are saying wait a minute
21:37you know a lot of these mega skills are
21:40actually moving away and they're going
21:41here or there so maybe the boogeyman
21:44isn't is bad but i yeah there's still a
21:46boogeyman I don't understand it
21:47right and so now they're starting to
21:49feel it out and sniff and understand a
21:51you know maybe there is maybe their
21:53applications warrant that kind of
21:55relationship with maybe not but now
21:57they're trying to explore and figure out
21:58what's going on so think think there's a
22:00lot more willingness now to try some new
22:02things based on that I think there's a
22:05lot more willingness to look at new
22:08shows up when you know Emily someone
22:10looks at it says you know what I need to
22:11try and like in our case we see you know
22:13some of the enterprises trying this just
22:16to make the incumbents kind of kick
22:18themselves in the head and and the the
22:20response is actually the most telling
22:21part right when when we're up against it
22:24in accumbens in a deal you know uh you
22:26know now we're still pretty young
22:27oftentimes the incumbent will come in
22:29and win but when they win they're
22:31winning on our message they're not
22:32winning on the message that they want to
22:34drive to the customer they're coming in
22:35and saying hey we can do everything
22:37cumulus can do and we can get close to
22:39their price points and you know you've
22:41been using us for years so just please
22:42keep using us yeah maybe one more just
22:45quickly to wrap up here the war between
22:49public and private cloud do you guys
22:51have any sort of thoughts on that sure
22:53so you know it's actually two years ago
22:57that you know there were a whole bunch
22:58of folks and this happens with new
23:00technology all the time because people
23:02just go to the extreme and there's it's
23:03all public right everybody is moving
23:06why bother and we're buying a disk and
23:09but the reality things settled somewhere
23:12in the middle and I think the argument
23:15now is not as much about whether it's
23:18going to be public private port for
23:20centage but the next generation admin
23:23the next generation admin is growing up
23:26thinking about cloud that's how they
23:28think I mean they grow up using you know
23:30iPhones they grow up pushing their you
23:32know pictures out somewhere to the cloud
23:34and being you know all that access and
23:36pay as you grow pay as you draw
23:38economics I mean they're they're
23:39inspired by these so the next generation
23:41data data center is going to look like a
23:44cloud whether it's on trim or or public
23:47and how what's the percentage of that
23:49depends on the industry how big the
23:51customer is on and so forth but every
23:53data center is going to look like a
23:56cloud that's a cloud inspired datacenter
23:58you know one of the observations that I
24:00have in it's just an observation trying
24:02to watch and understand is everybody
24:04that's been successful in offering up a
24:06public cloud you know the Amazons the
24:08Microsoft's now the googles they
24:10actually have a real business that pays
24:12for their infrastructure and I would
24:14venture to guess except for perhaps in
24:16the case of Amazon because they're so
24:17big right now if you got rid of their
24:19cloud component they'd still be
24:20profitable if you just try
24:21that off tomorrow said you get no cloud
24:23revenue there their whole right so this
24:25premise that public cloud is this
24:28ever-growing never ceasing resource I
24:30think is a bit flawed there has to be
24:32some underlying fundamental it allows a
24:34company to build that stuff out and if
24:37that's these thesis is true there is
24:39likely some math behind it that says the
24:42public part of your your infrastructure
24:44offering can only get to X percent of
24:46your total infrastructure and let's say
24:48for the sake of argument is 50 percent
24:49right so if it's 50 percent and
24:52Microsoft says ok I have X number of
24:54millions of computers that run 50
24:56percent of my business my real stuff
24:58then I can only have that same amount
25:00that I use for kind of generic public
25:02cloud offerings and most and I think I
25:05think there really is a ratio a golden
25:07ratio there that makes sense yeah
25:08interesting well thank you both
25:12Raman and jr. thanks for thanks for
25:15being part of this and of course we'll
25:17see what happens I'll let you know in 10
25:18years from now thanks mate