Interview: Gaikai's CTO Talks Technology, Lag, 'Gaikai-Powered' iPad Apps

Gaming on the "cloud" is supposed to solve some of the major problems associated with traditional earthbound methods of gaming. Cloud-based technology is poised to greatly reduce the need for expensive high-end hardware, curb piracy, circumvent the middle-man known as retail and provide greater accessibility to a publisher's titles. It will give gamers another viable option in how we play and obtain our games. At least that's what industry stalwart David Perry and his group at Gaikai hopes will happen. Along with OnLive, Gaikai is one of the two biggest hopes for cloud-based gaming in the industry today. Unlike Steve Perlman's OnLive, which launches server-based games through an OnLive portal, Gaikai -- which recently raised $5 million in venture capital -- will launch directly in the user's browser. A publisher can place an ad on a website for a game, and a user can click on that ad and start playing a game almost immediately, Gaikai says. Game publishers use Gaikai's streaming technology to allow users to play games directly from a publisher's site. Here, Gaikai's founder and co-CTO Andrew Gault fills us in on that technology, giving a broad overview of how it works and its origins. He also explains how Gaikai could grab a piece of the App Store pie with "Gaikai-powered" apps for individual games that are streamed to an iPad, and how lag won't be an issue. Gaikai will rely on datacenters -- at launch, how many data centers do you think Gaikai will need? How many would you anticipate having total? Andrew Gault: Actually, we could launch from just a single server. Of course, that’s not going to happen, but our business model lets us. Here’s how it works: you’re browsing the web, perhaps you’re on IGN.com looking at a review, or thinking of buying a game on GameStop.com or Amazon.com, it could be anywhere. Embedded into the webpage is some hidden Gaikai code that verifies your connection to our nearest datacenter and our current server load. If – and only if – we have the capacity and your connection guarantees a great experience we’ll show a “play now” banner. Clicking the banner will load up our client over the current web page and you’ll get to play a relevant game. You’ll never leave the page. The publisher pays out of their advertising budget, it’s free to the end user. If all our servers are full, or we detect you’re on dialup, no banner will appear. You won’t even know we ran a test. So with a single server in a single location we can run our "network" at 100 percent capacity 24/7, but we’ll be turning away tens of thousands of potential players every hour. In reality I expect to launch in about a dozen datacenters across North America and Europe to ensure coverage, with the servers un-evenly distributed across them based on population density. We’ll constantly be expanding our network coverage. Is the cost of having a large amount of datacenters a concern? Obviously if you need more datacenters, you likely have more business. But won't that get expensive? AG: We realized very early on that the success of Gaikai would depend on massive horizontal scaling. Each streaming server is completely self-contained – it receives IP connections directly from clients, it runs the games, it encodes the video streams, it caches game images locally, it maintains state with the rest of the cloud. No other infrastructure is required in a datacenter; we can literally rack a single server, switch it on and get a new location online. This lack of overhead was not easy to achieve, but makes it very, very cheap to add new datacenters. How long will it take for you to add new datacenters? AG: The great thing about our model is we see exactly where we need to expand. If we see that thousands of gamers are being turned away in Alaska because of high ping times, we know we need a new location up there. No guesswork, no fudging of numbers. At the end of every week we can run a report and see exactly where and how we should expand our network. As you can imagine this is perhaps the easiest possible conversation to have with our investors: "We turned away exactly 10,000 customers in Alaska and need exactly X dollars to support them." We’re already running this report right now, and we haven’t launched yet. That’s why I can’t be more specific on the number of datacenters at launch – it changes by the week. Gaikai was recently shown streaming World of Warcraft on iPad, which was all over the web. But while the tech may be there, would Apple allow something like Gaikai to run on iPad, considering the whole Apple vs. Flash conflict? AG: We’ve got Gaikai running on a lot of devices by now. Usually it’s just a bit of fun, but sometimes, like with the iPad, the experience just works. What you saw in the screenshots [Gaikai head David Perry] posted was not some Frankenstein hacked Flash, but a native iPad App. I don’t see any problem with getting Gaikai-powered Apps into the App Store. It’s a win-win scenario: users get great 3D games on their iPad and Apple finally gets to compete with the PC as a games platform. The key point here is "Gaikai-powered Apps." If we launched a generic Gaikai App that could play any game we deemed fit to stream now or in the future, we would effectively be bypassing the App Store. Not good. But if we provide a separate App for each game we stream, the App Store is still the central distribution point and we don’t stomp on anyone’s toes. Lag is the most obvious concern for gamers. I've read some of your and David Perry's comments on lag -- it is inherent even in locally played games, you guys have said -- and seen the Gaikai demo online. Are you convinced that lag is not an issue with Gaikai? What about when it goes live? AG: Every game has a different lag tolerance, both in terms of gameplay – think first-person shooter vs. turn based – and the target market – core vs. non-core gamer. Typically the "lag budget" for any game is within the range of 50 (Call of Duty) to 120 (World of Warcraft) milliseconds. (As a comparison, your flat-screen TV probably adds at least 50 milliseconds of delay, and your wireless controller another 100 milliseconds.) These budgets are well within the limits of our technology. Depending on resolution and client computer speed, we expect to add 10 to 25 milliseconds of delay to each frame, excluding network latency -- the ping. These figures are constantly improving. The one variable we can’t control is the network latency, but we can tune the thresholds in the hidden code on the web page. As an example, say a publisher wants to demo a new super-high-twitch title. We can ensure the "play now" banner only gets displayed if the ping time to our datacenter is less than 10 milliseconds, guaranteeing a total lag of less than 25 milliseconds. While this excludes many potential users, it makes sure that anyone who does play has a lag free experience. Put simply, we’d rather no experience to a bad experience. What's your biggest concern about Gaikai from a technical standpoint? Many might assume that it's reducing lag, but is that really the main concern? AG: The first major technical hurdle was lag, but that’s now in the past. It turns out it's relatively easy to come up with something that can capture and encode live video with 120 milliseconds delay (the threshold for a game like World of Warcraft). But getting that down to 10 milliseconds in order to give yourself some headroom for network latency is quite a bit harder. But what is really hard, is keeping the cost down. In order to make the numbers work we virtualize multiple game instances on each server. All the hardware is carefully selected and balanced, every piece of software from the kernel up is carefully tuned to try and squeeze another instance out. This is where most of our R&D is spent at the moment. We have the cost down to a price publishers are happy to pay as advertising, but we won’t be happy till we can get it down to a cent per gamer hour. At that price we can profoundly change not just the ease of access to games, but also the ongoing cost of gaming. David Perry has poked a bit of fun at OnLive's claim about being in "stealth" for seven years working on its tech. How long have Gaikai's engineers been working on this? AG: Rui [Pereira, Gaikai co-founder] and I built the core Gaikai technology for a slightly different purpose back in 2008, although the idea is a few years older. We wanted to create a YouTube-style portal for classic console games. We thought it might be possible to stream such low resolution video games – the SNES was 256x224 – in real-time, but we weren’t sure, and we didn’t even think about full resolution PC games. That artificial limitation stuck right through to the beginning of 2009. We were testing version one of the tech with friends and we got the now-obvious feedback: "This is cool and all, but I’d much rather be playing World of Warcraft." We’d spent months building an elaborate interface and control schemes to play multiplayer console games, complete with bells and whistles like webcams and chat, and we’d never even thought to give a modern PC game a try. Of course it only took about a day to get World of Warcraft running – and it was immediately obvious that it could work. Cue a complete change of focus. Since then David got involved, Gaikai Inc was officially founded and funded and we’ve built a great team. We’re just about to hit our 30th hire. Are you guys talking much more about the business model here? How exactly will Gaikai be making money? At least to begin with, we are a customer acquisition tool. We charge publishers by the minute to place demos all over the web, much like they would currently place advertising banners. The cost for a 30 minute demo is roughly the same as the cost of a Google click, except we aren't just driving traffic to a web site -- we're letting users experience the game. The publisher is free to decide how long the demo is, either in terms of play time or game progress. And it's completely free to the end user. Has the company decided on a launch date? No comment on a launch date, but I expect you'll hear more within a month or so. We're lucky, as the model doesn't require a huge marketing push to get users, we can just silently launch any day we think we're ready. All the necessary deals are in place.