YouTube's top engineer, wrangler of eons of video (Q&A)

What can you squeeze into a minute? YouTube uploads 4 days of video and streams about 3 months of it.

Venkat Panchapakesan, vice president and top engineer at Google's massive video site, is charged with not only keeping the machine humming but also figuring out how to stay ahead of new demands coming around the bend.

In his first interview since being named the site's tech head in June, Panchapakesan discussed the challenges of facing a future with ultrahigh 4K resolution; making YouTube accessible in emerging markets like India; and figuring out how to help users search the seemingly bottomless vault of video on the site. The following is an edited Q&A.

Q: Somebody told me an anecdote the other day that the Wi-Fi service on his flight warned him that streaming video sites like Netflix wouldn't work because of how much bandwidth they hog, but he was able to stream YouTube video just fine. He credited YouTube's technology for that; was he right?
Panchapakesan: It's a true testament to what we are trying to achieve. I really feel good about that...I've been here a few months, and there's definitely been ramping-up and learning. But when I look at YouTube and say, what are some of the big goals, the big challenges -- the goal for us here is to keep up where technology is heading, but in some ways get ahead: the cameras, the video-capture technology are going to move very, very fast. How do you deliver that rich experience to people regardless of which device they are on, and on whatever bandwidth they are on.

Q: What are some of the measures YouTube has taken to reach that goal?
Panchapakesan: We saw a huge improvement in HD quality that you can deliver for half the bandwidth that is normally used -- people are spending more time watching. They feel like they're not looking at a streaming video. It's the same thing that we have done in ExoPlayer [a media player for Google's Android mobile device operation system] that we have launched. We want people to have this experience whether they are on Internet Explorer or Safari or iOS -- to be able to have the best video experience possible.

Google and YouTube are uniquely positioned to solve these problems. We are seeing a lot of that traffic, we're seeing a lot of that usage, and we're also seeing the trends that are happening in the technology, particularly in the camera and video-capture technology. We want to be ahead of that.

Q: In getting ahead of the curve, why do you think 4K is something people will want to watch with YouTube? Some are unconvinced that YouTube's content will be the sort of thing people will want to watch at such high definition.
Panchapakesan: The human eye -- obviously, we enjoy what we see in nature. And when you think about the TV screen, the evolution of screens in general, you can see the difference between what you are experiencing in real life and what you watch on a screen. People were using DVDs; then when Blu-Ray came, the reaction was "Oh my god, I don't want to watch DVD anymore." The same thing is happening with HD, it's a natural human reaction to things. We all want it to look like reality.

Q: When viewers get buffering, hangups, quality drops and time-outs, should they blame you? Peering complications? Their ISP? Who should they be frustrated with?
Panchapakesan: It's a fair question. We want to give transparency to the user, so we launched Google Video Quality Report. It lets users in their homes figure out, why am I seeing this problem?

But the way that I'm approaching this is: We can do a lot to make this experience better -- lead time before rebuffers and buffer rates. How can we reduce that through technology? Can we do better encryption/decryption? How would we stream the thing more effectively? If you're watching a video and focusing on one part of it, maybe that part can be sent at a higher resolution. We are starting to play with what would make sense. Our approach here is that it's not something we need to do only for Google -- I think these things are going to benefit the broader video community. We want to work with other companies in the space.

Q: Are the challenges different in trying to make YouTube accessible in emerging markets?
Panchapakesan: I travel around the world a lot, for the job and also personally. You look at stats today, half the Internet population is already on YouTube, and we need to make things better for them. But then there are going to be a billion to 2 billion more people in the next year or two who are going to have access to devices that allow them to watch video. Infrastructure and cost are the two big factors.

So one of the things we just did a couple weeks back, we had this event in India around Android One where you can offline YouTube videos in India: if you are in a Wi-Fi area at one point, you're able to go home and you can still watch the video. So we're experimenting with how do you deal with markets where people's devices don't have enough data, and what can we do to address those needs.

We're also working with some carriers to do data plans. We did one in Malaysia and two in India. They're giving users an introductory offer, so they can watch YouTube for X amount of dollars -- you get so many megabytes. Very, very cheap plans in some cases, lower than what people would typically pay.

Q: Searching YouTube, what do you need to improve? What will it mean to have Google's search guru Udi Manber in charge of YouTube search now?
Panchapakesan: If you think about YouTube, one of the things that completely amazed me is the scale. Every single day we get lifetimes worth of video being uploaded. Figuring out in the oceans of data that people are uploading: How do you get the right set of things that people can see?

There are two aspects of this -- one is search; people can search. These are not documents, right? There's video, there's audio. So how do you figure out what is the right video to show? Second is people come and they want to browse. They watch a video on YouTube and we give them suggestions for what would be the other videos they would be interested in. Which people love a lot, but when you have such a small real estate to show them -- let's say, six videos that you could show -- what is the thing that this person wants to watch next?

And that is a huge challenge, so we do a lot of machine learning and understanding of user behavior. We are starting to explore some deep learning techniques, which are neural-net-based learning models that I think will be very, very helpful. Those are things that Udi is starting to look at. Obviously, Udi was on Google search, he has a lot of context on what to do in search and he's been in that space for a long time. But I think this notion of delighting every single user, giving them the things that they are excited about -- is it about context or search context -- it is a fascinating. A lot of good work happening there. And I think we want to continue to invest.

Correction, 5:40 a.m. PT: This story initially misspelled the last name of Venkat Panchapakesan, vice president and top engineer at YouTube.