WordPress was built to be global scale from the beginning, says Mullenweg

[protected-iframe id=”823a00083553904d610cc47121ccebfe-14960843-25766478″ info=”http://new.livestream.com/accounts/74987/events/2361507/videos/30327079/player?autoPlay=false&height=360&mute=false&width=640″ width=”640″ height=”360″ frameborder=”0″ scrolling=”no”]
Session Name: What Does It Take To Run 19% Of The Web’s Content?

Chris Albrecht
Om Malik

Matt Mullenweg

Chris Albrecht 00:05

Thank you Mr. Bell for making my job as MC fell very very very small. Coming up next, I’m going to welcome to the stage my boss, Mr. Om Malik. He’s going to be in discussion with Matt Mullenweg the Co-Founder of WordPress and Founder of Automattic. I’m talking about, what does it take to run 19% of the web’s content? Please welcome to the stage Om and Matt.

Matt Mullenweg 00:32

Thank you, thank you. Good morning everybody.

Om Malik 00:37

You should liven up the proceedings.

Matt Mullenweg 00:40


Om Malik 00:40

Let’s talk about– it’s been a–

Matt Mullenweg 00:41

Do you know the macarena?

Om Malik 00:42

— busy week for you already and it’s not even Wednesday. What’s going on dude?

Matt Mullenweg 00:46

It is Wednesday, right?

Om Malik 00:48

It is.

Matt Mullenweg 00:48

Okay, okay, just checking.

Om Malik 00:50


Matt Mullenweg 00:50

The time-zones trip me up sometimes. It has been… yes, a busy week. Yesterday, we announced that Tiger Global Management has put an additional $75 million, in addition to the 50 that went in, in May, in the secondary stock in Automattic.

Om Malik 01:08

So that’s a big company now.

Matt Mullenweg 01:10

It’s a big company. We are getting to be a bigger company. We’re about 195, or probably be 200 by the– 204 by the end of the month. So, of that 80 who have joined just this year, so it’s been definitely a period of rapid growth for the company [which is interesting?], because for many years I was trying to keep the whole thing under 20 people.

Om Malik 01:33

How did that work out?

Matt Mullenweg 01:35

It didn’t work out when I [?].

Om Malik 01:37

I know because–

Matt Mullenweg 01:38

As a matter of fact there were 10.

Om Malik 01:38

— I remember– I remember suffering through the [?]. No, jokes aside, we’ve been users of WordPress.com from the day it started and it is probably one of the best Cloud services out there, and one of the reasons I wanted Matt to come up on stage is to ask him what it takes to build a Cloud scale company, a start-up, a Cloud scale web start-up [with?] essentially very little money about $20 million in its [?] until till recently, and power almost 19% of the web, so let’s talk about that. How do you get to be a company, which is– doesn’t exist one day and four year, five years later you are bigger that Yahoo on [Quantcast?] ranking.

Matt Mullenweg 02:35

It’s– we were very lucky early on. We started sort of before the golden age of Amazon and all the other services, and it was just really [?] end of infrastructure and service were you could go on– what was the one before Rackspace? The little one in Houston. Or this little server [come easy?], put your credit card in and you’d have a server 10 minutes later and that was kind of a physical server somewhere, and that was kind of our equivalent. And early on we chose a bad one, and we had to go to two day [listeners?] simultaneously; we only had 10 servers. But I couldn’t afford to move all 10 severs, so we had like eight in one day [listener?] and four in the other day [listener?]; we sort of transition just slowly. Now, the good news about that is we wrote all the code and debug all the problems about doing multi-day [listener?] serving [just to save?] life because we have a server at life from both. So, cashing, working across data centers, all those sorts of things, so we did that when we were very, very small. So as we grew to thousands and thousands of servers across and be a presence in eight or 10 data centers now, we just scaled up some of that code we had done fairly on and it worked great. You almost never ever hear about WordPress.com being down, and when it is, all of your blogs are offline, so you don’t hear about it.

Om Malik 03:58

Have you heard of this thing called Twitter? Just letting you know. So, how many servers do you guys have now?

Matt Mullenweg 04:06

I do not know exactly, it’s in a thousand.

Om Malik 04:09

Like tens of thousands or a hundreds of thousands?

Matt Mullenweg 04:12

Ones of thousands.

Om Malik 04:14


Matt Mullenweg 04:15

There’s been a lot of gains and efficiency that we found over the years. So, for a while, while we were growing very [linearly?] in severs, Moore’s law and other things that we figured out, basically figuring out inefficiency in the code has allowed us to sometimes some [?] even drop the number and have more performance and more [?] and everything.

Om Malik 04:36

So when you were writing the code in the early days of WordPress.com, what were you guys thinking, that you will be at in a tens of thousands of servers, or what was the [start?] process?

Matt Mullenweg 04:50

I think very much so, because one of the selling points of us hosting your WordPress versus you doing it yourself was that it’d be totally bulletproof. And that– do you remember the [Dig effect?]?

Om Malik 05:02

No. Nobody remembers Dig.

Matt Mullenweg 05:04

No one remembers Dig. It was like [Reddit?], so it’s in the [?] [bacteria?] site at once, and sites would always go down, and we want to be Dig proof which we did. We even now survive the whole company.

Om Malik 05:19

Right. But was it like part of the decisions making process that the resilience and scale, like a massive scale, was going to be part of the software code? I mean, not many start-ups think about, like they’re going to become 19% of the web or 20% of the web, right? So, you don’t really think about building out a big Cloud infrastructure?

Matt Mullenweg 05:44

I think that from the begging we were founded somewhat on the assumption that blogs would replace all of the traditional media in the world. And WordPress as software could be flexible enough to power– just the other week New York Post switch to a 100% of the site over [?], or catching up to GigaOM eight years ago.

Om Malik 06:06

That’s good.

Matt Mullenweg 06:07

And when I post and pay checks, and all of those things are 100% WordPress. So that was the thesis, and so… yeah, from the beginning we thought this needs to work at a higher scale. And that’s worked pretty well, and it’s been good to be ahead of the curve there because a lot of the early adopters of WordPress were media. New media and old media, and that’s driven a lot of the more main stream adoption, because people say, “Oh, they blog on what they read.”

Om Malik 06:31

So what is the big take away? What’s the big learning for the Cloud audience from WordPress.com? Like, what are the things you can share with the crowd here, which they should be thinking about up-line to their business and infrastructure?

Matt Mullenweg 06:46

One thing that we did from early on that I think worked well was, I’d always tell the engineers, and as an engineer myself for the Automattic. I always said, “Just assume that CPU and bandwidth is infinite.” And then I tell our system folks, “Hey, hurry up and make it so the engineers don’t notice that it’s not.” So, there was always that kind of constant balance and tension between us launching services which sort of– because otherwise, what’s the point of having a Cloud servers? You can host WordPress with a one-click install out of every $5 GoDaddy and Bluehost in the world, so what’s the point putting on us? So we did very early– this year we launched the service called [Photon?]. And basically it takes on Jetpack powered blog, which is sort of a way to hook your self-host to blog up to WordPress.com. It takes all of your images and dynamically re-sizes them and puts them behind the CD and everything like that. So, effectively, we became a free CDN for a significant portion of the internet overnight, and that was exiting, but that was we wouldn’t have built. We build that for WordPress.com first, because we thought, Why do we need a pre-generate all this sizes of images, let’s just have one golden copy and make everything else on the fly. Kind of like a, everything is a shadows on the wall of Plato’s cave.

Om Malik 08:04

Right. And so, you’ve done the same for video as well? I think you guys host videos too.

Matt Mullenweg 08:09

Absolutely, trans-coding videos and other high CPU task, but the– I think what we’ve done differently than a lot of Cloud services is you can have a choice. So, with WordPress it’s fairly unique in that– you can– as a service, WordPress.com, that you can sign up for– pay anywhere from $0 a year to tens of thousands of dollars per year, so it scales up and down like that. But if you want, and I think it’s the only Cloud service like this at scale, you can take you WordPress data and run you own WordPress with the exact same code that we’re running. So, the entire stack is opensource, all the way down to the plugins we use for multi-data center database replication and things like that. So, that’s fairly unique and that you have complete flexibility, but then we created this thing called “Jetpack”. Basically, what Jetpack does, is that if you’re running it yourself, it brings in all the goodies, all the services, like the video transcoding, photon, connections to social networks that sort of are– or built into a singular connection [?] like Facebook, Google+, and brings it to your self-host to blog; things that you’d honestly really can do yourself without writing a lot of code or doing deals with Google.

Om Malik 09:22

Right. Let me ask you a little controversial question, In a sense that you use PHP and MySQL, like two technologies which of, kind of, fallen out favor with the new generation companies, developers and… And also, the Cloud people don’t talk about them very much. So, what’s your experience with those two languages and technology? The database technology.

Matt Mullenweg 09:54

I’m a big believer in the right tool for the job. PHP is actually pretty incredible as a workhorse. We’re not the largest PHPs in the world. There’s Yahoo, Facebook, Wikipedia, there’s a lot of folks using PHP as [?]– [we’re?] still pushing the language for, but again, it’s a right tool for the job. So, for a number of systems, we have some real time stuff [?] written Python. We have [yearling?] stacks for [Java?] Messaging Service, and we’re doing more and more with Node is as the need arises. The beautiful thing is that, the web is such a beautiful transport layer for connection [?] services, it doesn’t really matter what they’re built on. The core of WordPress will probably always be on PHP and MySQL, or Percona, or MariaDB; MySQL compatible thing, because that’s what runs everywhere. Like I said, every single one of those $5 [?] web-host in the world are on PHP, and Amazon, and Google app, and everywhere. It’s like the [“lingua franca?”] of the web. But for higher, more– not, propriety is not a good word. For services that don’t need to be distributed in the same way. Sure, you can use whatever you r engineers most want to use; we’re not religious about it.

Om Malik 11:09

Right. So when you look out– like after learning for about the last eight years of doing this. What do you look at when you look into the future? What do you see the future of the Cloud look like? What do you see web services– how do you see them evolving?

Matt Mullenweg 11:27

I’m most interested in things that Cloud services can do that would just be almost silly to contemplate in an on-Cloud world. And I think, particularly, as we start to get more Cloud services, actually, provide more vertical scalability rather than just being, sort of a, an API layer on a virtualization. That’s pretty interesting.

Om Malik 11:50

Can you elaborate a little bit on what kind of service you [?].

Matt Mullenweg 11:54

Sure. So, even on Amazon, easy too today. We choose like a size of instance making sure you’re getting a virtualized amount of resources, with [fine?] amount of memory and CPU into space, etc., versus [Matt Mullenweg?] where everything is obstructed, and you just get a– you can put as much data in there as you want, basically. Or WordPress.com where you can send as much traffic to GigaOm.com as you want, and it just sort of dynamically adapts in the background across data centers, across CPUs, across everything; all the scaling’s invisible to you as a developer on WordPress.com. So, those type of services, I think, enable the creativity and sort of a friction free ability to try newer things. Like for example, how many of the [?], including you guys, do live blogs now? It’s not something you would ever probably comprehend in a machine world, because the way we deal with [some of our?] back-end might spread live ball of traffic across hundreds of servers, that you would only need one day a year.

Om Malik 12:54

Right. So, you’re saying that, from the point view of an application or like a service, the limitless nature of scale should open up more opportunities and more freedom to tinker and think from a [?]. What about this emerging [?]– you and I have talked about this privately quite a few times. The idea of data informed applications, do you see that becoming part of all applications in the future? Do you think– one of the things we argue about is that the need for– especially on the mobile phones, the applications to be intuitive and smart, and using data to just create a better experience. Do you see that actually taking a whole or is it something I’m imagining?

Matt Mullenweg 13:46

Yes, but it’s– I feel we’re actually at a low point right now for this, because our ability to collect fast amounts of data has outpaced out ability to use it effectively and analyze it and really understand it. I think it is most– actually, from the point of view of creating software, creating an application, you can [?] as many inputs as possible to know how people are really using what you’ve built, and to tweak and iterate on that, and so we do– at any given time, WordPress.com has probably 15 or 20 multi-variant test running and we’re cohorting users into different sections in doing a AB test, so you know how multiple controls and things like that. But, it’s tricky to build an organization for anyone whose building or start about there that doesn’t become a slave to this, and the data can be misdirecting as often it is directing. So, sometimes you’ll see a number that goes up and up and up, and if you don’t really dig into that and know the why, why it’s [coming?], you could assume you’re doing a good job. And so, one day you start looking at some of those things being [credited?] and you just see, “Oh, spammers figure out how to [descript?] our signup,” and you need to get rid of 70% of your signups now, or that you need to build systems that adapt to this. There’s lot of examples of that that I think it can be– the data could be misleading. So, what we’re trying to do is create a culture where people are data informed but not data driven in creating our products, and I hope that we can create the user experience user data to be the same way. So, I prefer, call it Twitter model of following than a Facebook model, where they try to figure out everything [?].

Om Malik 15:29

Right. You know, let’s talk about that. I think there’s an interesting– some of us, old time, where people like me, I remember Yahoo in 1999 and 2000 laughing MyYahoo and – what was the first [?] personalization. And then the [boost?] happen and personalization never really arrived, and it still hasn’t really arrived. Like the web is still pretty– it’s not really highly personalized the way you would have expected. And now, given that you just passed Yahoo as on the [Quantcast?] list, either maybe you have some idea on that.

Matt Mullenweg 16:10

I think that all the services that have been most successful have been sort of hyper person [honest?]. I mean, what’s one of the most visited pages on the internet, the Facebook New Feed, because I think a remarkable [?] personalization. You can’t– I use to have this thing where my business card just said: “Go to Google, type in Matt, press I’m feeling luck,” because for a while I was [?] head on Google for Matt. But now there is as many versions of the search results for Matt as there are people in the room. Google looks at you search history and your email, and etc., etc., and personalizes it. So, clearly the most– one of the most complex task on the internet which is search of it, is now 100% personalized. And I think that is a– but when it works, it’s invisible. So, versus the Yahoo model where the personalization was really, you just picking things and moving them around. Our ability to gather data and for the best in the world, like the Facebook’s and Google’s to analyze it, allows them to create sort of a invisibly personalized experience. And I think it’s very compelling, but also probably has some implications for– perhaps in [?] especially. Your viewpoint becoming [slow?], and perhaps seeing too much of the things that you already agree with and already like.

Om Malik 17:27

Is that a line you’re using, why there is no personalization on WordPress.com just yet?

Matt Mullenweg 17:32

We personalize it, you just don’t know.

Om Malik 17:34

I think so. I don’t think so, but I think the Facebook personalization doesn’t really work, because I frankly do not want to congratulate people on getting married a millions times and baby pictures because I have no interest in other people’s kids.


Om Malik 17:51

[inaudible] I just don’t. It’s like– and they keep showing me the same thing, and this like– wait, I liked somebody’s picture a year ago and now I’m–

Matt Mullenweg 18:00

Before the [kid?].

Om Malik 18:00

— stuck. No, when they had a kid, and now I’m stuck. Basically, everybody’s kids are being shown to me; I don’t care. I’m not a [?], kids are great, they keep the population growing and stuff.

Matt Mullenweg 18:14

You said two things that are wrong there. First, that you’re not a comedian.

Om Malik 18:18

I’m not.

Matt Mullenweg 18:18

Not, and probably not true. And two, Facebook actually has some pretty amazing customization in the field, you should dig into it. Like person by person, you can choose what kind of updates to see, what types of events to see? My favorite is, is seeing the relationships status change. Those are often the most interesting things on Facebook.

Om Malik 18:35

Really? I’m going to follow your relationship status. I’d never notice that but I will do it now. Thank you Matt for making the–

Matt Mullenweg 18:42

It’s been a pleasure.

Om Malik 18:42

— time for us, and good luck with everything.

Matt Mullenweg 18:45

And thank you guys.