After the debacle of yesterday’s on-stage interview with Mark Zuckerberg at South By Southwest, I decided to take myself out of the process as much as possible by offering up a Q&A. The results below are heavily edited excerpts from my conversation with the Facebook CEO. Instead of grilling him too much about monetization and the company’s $15 billion valuation, we spent most of the time talking about infrastructure, scaling Facebook’s architecture and what the ad platform is going to look like. And I’ll tell you right now: When I asked about money, my questions were quickly shot down.
ME: What does the Facebook infrastructure look like right now as it supports more than 67 million users?
ZUCKERBERG: I can’t talk about the number of servers specifically, but I can say it’s not on the order of hundreds of thousands yet; it’s in the order of tens of thousands. Would you like to talk about how the architecture has evolved over time?
It’s pretty fascinating to see that the evolution of the network architecture pretty much reflects the evolution of the site. When we were in a few colleges one of the theories there was that most of the people would want to talk to people at their own schools, so we set up a network architecture based on the school. We had a Harvard database and web server and routed traffic to them through sub-domains. As time went on, and we reached a reasonable scale, we had to break out into tiered services.
In mid-2005 or so we also started relying heavily on memcached, and today it’s become such an important part of our infrastructure that we’ve actually become the head developers on one of the branches of the code.
At this point in Facebook we’re now running our networks and have multiple data centers across the country, one on the West Coast and one on the East Coast. So we’re working on problems of reducing latency in transferring data from one to the other.
ME: So as you expand internationally will Facebook need an international data center? When?
ZUCKERBERG: That will be something at some point.
ME: Do the applications running on Facebook affect the network? How much of that does Facebook need to support?
ZUCKERBERG: We have an API tier that serves the data requests, but the same servers serve the requests generated inside of Facebook and those form outside of Facebook. We’re trying to blur the boundaries of content created inside Facebook and applications created outside Facebook, and that’s reflected in the architecture.
ME: What challenges will you guys need to address as you scale?
ZUCKERBERG: We put a ton of focus into making sure going forward that our development speed will be very fast. But you know the phrase, “Premature optimization is the root of all evil”? If we spend our time trying to optimize our architecture it could end up being one that’s difficult to develop on and going forward, a big focus for us is making sure that’s not the case.
One of the recent places is internationalization. The infrastructure to do that is pretty fascinating but it has the potential to slow down future development as people ask if they need to develop an application in another language first, so we’ve basically built a wrapper for every string of text and the developer puts a wrapper around that text and we send it to users to translate. So that doesn’t slow down the development speed.
ME: How do you handle quality control?
ZUCKERBERG: In the three cases we’ve done so far (Spanish, German and French), the quality of the translations that we got was actually better. If you wanted to get professional translators it wouldn’t be scalable.
ME: Let’s talk about monetization. You said yesterday that you envision the social advertising landscape evolving over the next 10…15…20 years. How will those ads evolve and when will we start seeing aspects of them on Facebook? And where does Beacon, which you said wasn’t an ad effort, fall into this?
ZUCKERBERG: Beacon was a part of the platform. It was part of this while effort to blur the boundaries between what’s inside Facebook and what’s outside Facebook. Beacon was our first cut at a protocol to do that.
When it comes to social ads we really want to line up what people are trying to do on Facebook and the utility it offers with monetization. If you look at what people are trying to do on the site, it’s communicating and connecting with each other and sharing information, so the business model should be around people sharing information and staying connected.
In banner advertising, people who have developed a trust with the audience run a banner ad and the trust bleeds over to the ad so people pay attention to it.
ME:(making a skeptical face.)
ZUCKERBERG: What? You look like something’s wrong? People go to a content site to see a specific kind of content and will trust those ads relate somehow to it. On Facebook, people aren’t coming to see content from Facebook; they’re coming to see what other people are sharing, so the most natural analog would be having the ads be information shared among the people. Because so much of our society has some commercial component it seems like there will be a way to both share information and line that up with what advertisers want.
Some amount is happening as advertisers pay to accelerate that distribution of information. The amount they’d be willing to pay is proportional to how much it is accelerated.
ME: And before we leave I wanted to ask about revenue and the fact that you mentioned yesterday that you are operating at breakeven.
ZUCKERBERG: I said we’re operating around breakeven. In the past we’ve run at sometimes more and sometimes less, but as a private company we don’t disclose those numbers. A lot of people ask us what we’re going to do with that Microsoft money, but we have no specific purpose for it. We didn’t take that money to buy something.