Monday, March 30, 2009

OnLive: Behind the Scenes

My friend--let's call him "Kojak"--shared some interesting details about his employer, OnLive, who made a big splash last week at GDC (read more about it: company web site, press conference video, Penny Arcade comic).  Some highlights from his email to me:

Still recovering from several days of stress and almost no sleep, but I'm relieved that our big announcement is over. I wish it had been all smoke and mirrors, but unfortunately it was all live and relied on servers in our Santa Clara datacenter.  This necessitated insane fallback options, including the "Tertiary Contingency," which I can't really go into.  Let's just say we're all very grateful it wasn't necessary, leaving us instead worrying how far Steve [Perlman] would stray from the script.

Most of these facts have been mentioned elsewhere, but some journalists missed the significant bits, so to speak:


This is of course the big one that everyone's talking about.  A couple of early news articles misinterpreted what Steve said in previous interviews.  All the bloggers picked this up as gospel truth, and distorted it further.  On the morning before the press conference, almost every mention I saw had our total round trip latency being <1ms. Anyone with half a clue pointed out this is completely impossible, and it led many to assume we're just another in a long line of charlatans.

Steve tried to clear this up a bit in the announcement, saying that "encoding latency" means what we add by running it through our proprietary encoding card in the datacenter (this is really our key technology). "Last mile latency" can add anywhere from 5 to 25ms, depending on your ISP and other factors. Improvements in that in recent years are key to making this all possible. The latency from Moscone to our Santa Clara datacenter ~50 miles away is <2ms (I wouldn't have thought that possible a few years ago).

He's also talked about the contribution other sources of latency make, in particular your display. Gamers who switched from CRT to LCD may have added more latency to their experience that we could ever add (assuming worst case LCD and best case network). As you probably know, LCD TVs can be even worse, especially for those who don't know how to turn off the various "pixel shining" features. We literally only found one model of Sharp LCD TV that was really optimal for games (and apparently we bought every one in northern CA). Many people happily play Halo 3 on some of the worst latency TVs without ever knowing the difference. What's great for movies isn't necessarily great for games. Don't even get me started about the latency added by most wireless controllers.

It's expected that certain titles might not really be playable due to latency, but I've been pretty surprised so far (we've all been forced to play lots and lots of games at work!). Mirror's Edge was one I really didn't figure would work, but it's done pretty well.

Bandwidth caps

This is probably the most valid concern I've seen people raise, besides those who don't think they can get a consistent 4-5Mbps in their area from any ISP.  I don't know what the official Comcast monthly limit is in most areas, but Steve has been saying 250GB.  Since you're rarely pushing the peak bandwidth, average use is much less (and varies widely between games and play style, even if everything's running at 720p60). If it's ~2Mb, that gives you 278 hours or 11.6 days of continuous game time per month (assuming you're not using the connection for much else).

We expected the major ISPs to be pretty hostile, but they ended up almost scaring us with their enthusiasm. As long as we're causing a predictable load and we're willing to peer closely with their networks, they don't seem overly concerned (and Steve talked about various bundling possibilities here and in other interviews). So, it's realistic to expect either a special "OnLive" tier from your ISP with no cap, or some other arrangement where our data usage doesn't contribute towards your cap.

That doesn't mean it won't be a rocky upgrade path for some of them.  If this takes off like we hope, they're going to be very busy, and it won't work perfectly in all areas for a while. This is part of the purpose of the Beta.

Target audience

We're targeting all segments of game players, but we don't expect we'll ever satisfy the most discriminating/insane gamers (the ones who shell out $5K/year for a top-of-the-line beast to play first person shooters at one frame/sec faster than their friends).  However, by leading most of our demos with one of the most demanding FPS out there (Crysis), we try to make it clear that we're "good enough" for most who would want to play even the most extreme titles.

We have hopes of eventually combining our MOVA facial capture technology with customized servers to allow experiences beyond anything a console or PC game could offer.  We'd certainly give anyone willing to do an exclusive title access to some interesting technology, though I'm not sure who would buy "Benjamin Button: The Video Game."

As with WebTV and other resource-intensive services, our ideal customer is someone who pays their bill and hardly ever logs in.  Some have said we're ideal for the mythical "casual hard-core gamer"--someone who likes to play the latest high-end games, but only a few hours a month, and thus can't really justify maintaining a PC capable of playing them.

We're also really interested in true casual gamers (people who enjoy Xbox Live Arcade titles more than any of the $60 ones from the store).  As Steve mentions, the hope here is that many of these titles will eventually be able to be "virtualized". This isn't quite as far fetched as it sounds, as Xen already has experimental support for GPU virtualization (so one video card could really be shared between multiple users).

Though Steve touted that we were demonstrating "every type of game from all major publishers," you'll notice one obvious omission: no MMOs. This is not a coincidence. Most of these games would be difficult to virtualize (they're not Crysis, but some are getting closer), and MMO gamers neither sleep nor work, apparently. So even though MMO games work great on our service for several reasons (slightly less demanding of resources and their users are accustomed to poor service), it's hard to imagine any business case that would make sense. Sadly, many of the bloggers that seem most enthusiastic about our service hope to use it exclusively to play MMOs. It could happen, but don't hold your breath.


We're not a "streaming" service.  Apparently voice actors now insist on certain clauses in their contracts that refer to streaming.  Thus, we do not "stream" games, but instead provide them in real time, over the Internet. It's easy to see how there might be some confusion here.

Meanwhile, at the demo booth...

Highly skeptical gamer shows up and plays for quite a while. Finally, he says, "I'm an Atheist, but I feel like I've just seen the face of God!"

Someone from the Xbox Live group at Microsoft drops by our booth.  After briefly playing a game, he quickly makes a cell phone call and says, "It works." Then he dashes off.

Thanks, Kojak!

D and I have both signed up to be beta testers, because we may actually be examples of that mythical casual hard-core gamer.  Before buying her last two computers, D literally checked their specs against the system requirements for the current version of The Sims.  And I maxed out my laptop's graphics, memory, and processor speed when I bought it last January so I could use it to play Spore.

I remain skeptical about OnLive, given that we still have trouble with streaming video (Hulu, Netflix, etc.) over a supposed 12Mbps connection, but I'm willing to give them the benefit of the doubt.  I know several people who work there, and they're all pretty smart cookies.  If anyone can make this crazy scheme work, it'll be them.

No comments: