Slashdot Mirror
Cringely on P2P vs Streaming Data Centers
Anonymous Coward writes "Robert X Cringely is postulating today that as bandwidth applications grow, the data centers will never be ready to serve 30 million concurrent streams of data. Akamai, with its tens of thousands of servers spread in an intelligent topology, still can't serve more than 150,000 concurrent streams, which is never going to impress the TV network exec used to audiences in the millions. Cringely choruses that secure P2P is the solution to delivering not only high quality video but also to audiences that scale in the millions. BitTorrent seems
to have worn out it's welcome with the MPAA recently, so maybe the future holds P2P networks owned and managed by Hollywood?"
Sure, currently 150,000 copies of data puts a large strain on the servers... what about one copy broadcast via multicast, much akin to airwaves?
... multicast and proxy technology that we have spent the last 10+ years working on to solve this problem?
"Akamai, with its tens of thousands of servers spread in an intelligent topology, still can't serve more than 150,000 concurrent streams"
Assuming Akamai has only 10,000 servers, that's 15 streams per server. C'mon now, we're not that stupid.
I'll subscribe to Slashdot when I see a month without a dupe, a typo, or an article the "editors" didn't read.
Content creators and content consumers are becoming one and the same. You can see this every day on sites like Jamendo and Flickr.
Great. Another prediction on what technology will or will not be able to do in the near future.
We all know how accurate these are.
Also: There is a difference between serving the exact same fucking content, at the same time to 1 million people and generating custom pages on-demand for 1 million people.
I think the key issue is that everyone is asking for flawless, high quality, on demand data streams. We currently have streaming broadcasts over a User Datagram Protocol.
The difference between UDP and other protocols is that UDP does not ensure that packets are not lost. This works well for audio and video because if you miss a frame or two, you probably won't notice too much. This is the equivalent of broadcasting a signal over the air waves. Sometimes it'll be a little fuzzy, but you can still understand what's being sent.
But like I mentioned before, UDP streaming broadcasts will not give you a high quality, 100% accurate and on demand data stream. That's why we're focussing on P2P instead.
I think the worries over the datacenters is a bit unfounded at the moment. 10 years ago, using 1GB of harddrive space and ever needing more than a 14.4kbps modem seemed insane. But now things are different. And the cable tv to internet swtich won't happen overnight. I think our technology will catch up by the time it catches on.
Capitalism: When it uses the carrot, it's called democracy. When it uses the stick, it's called fascism.
The Akamai figures are the embellishment of the submitter... Cringely doesn't mention Akamai anywhere in the article.
And thus I don't really think they will switch to this model. Simply put: Their "servers" would not be under their control. If we were to provide them with "servers", we could at least partly control what is shown.
... use your imagination.
Of course we would not get a say what we distribute. But that's not the point. You cannot rely on a P2P Server to provide real time content. Suddenly it's gone, because I switch the box off. Even if you have a few fallback "servers" on the list it's nothing you can build a reliable service on. And people do get angry if their favorite soap suddenly skips right after the words "I kept silent 'til now, but now I have to say it. I am..."
Not to mention the danger of tampering with the content. Yes, they will encrypt it, yes, they will make it near impossible to inject anything, but there is still the danger that in the middle of a Disney Movie you suddenly get to see
We used to have a Bill of Rights. Now, with the rights gone, all we have left is the bill.
Whatever happened to the MBONE? I see that a book on the subject is now posted to the web and freely copyable because it's gone out of print. The MBONE FAQ dates from 1993. That's like (/me whips out his HP-41C calculator) 13 years old. Apparently the IETF has a group for MBONE Deployment, but it hasn't been updated since last September, and even then it was a year late for its final milestone.
Don't piss off The Angry Economist
Chaincast
NetCableTV
Red Swoosh
Kontiki
Just to name a few.
Some of these have been in production for many years. Chaincast is/was the leader in radio streaming (at one time).
There are more advantages with P2P streaming/downloads than meet the eye. You also get better sharing of data in the local network. i.e. you're at Starbucks, you see someone watching somthing you want too - start the download an you get it at full speed from one laptop directly to the next. Also, from an infrastructure pespective, it's automatically fault tolerant.
It's big.
Multicast has been deployed on Internet2 for some time now. I've watched 720p streams multicasted from Europe with no problem.
The problem with deploying it on the commercial Internet is political. Backbone commercial Internet providers have had multicast on for a LONG time. ISP's that give you your home broadband connection which are mostly cable TV operators and companies like verizon don't want to provide a cost effective way for content providers on the net to deliver video. They would rather charge you for their "middleman" service. It's not like they don't know how to enable it, all they need to do is enable it on their switches and routers.
Most cable operators use multicast already to stream the channels through their set top boxes.
In Britain The BBC is working with ISP's to multicast to broadband connections. That would REALLY be nice if something similar happened here (In the U.S.)
http://www.bbc.co.uk/multicast/
Cringeley doesn't mention Akamai. Where does this 150K max users figure come from? If "tens of thousands" of servers is only 10K servers, then 150K streams is only 15 streams:server.
But even a $2K P4/4.3GHz can serve over 1750 simultaneous 500Kbps video streams (from my own benchmarks), for 875Mbps. Since Gbps fiberoptics cost <$5000:mo, or under $3:stream:mo, 10K servers should serve at least 17 million simultaneous users; 58K servers serve over 100 million simultaneous streams.
Use more efficient servers, like SANs coupled more directly to routers, and you're talking about <$3:stream:mo for maybe 100K servers serving over 1 billion people, for a $100M investment that can be amortized over a few years. Years which can bring maybe $1-100:mo profit on 1-10 billion consumers, or 10-10,000x ROI.
Such a network is much more efficient and economical as P2P, or multicast. But even the raw numbers sound very profitable. That's why Akamai is making so much money, even though their market is still so small.
--
make install -not war
IPv4 multicast across the Internet will never happen.
The reason is the complexity involved in deployment (multiple protocols, MBGP, MSDP, etc.) and that you have the 'third-party problem'. Basically both transmitters and receivers have to rely on a third-party for a redezvous-point.
Scalable Internet wide multicast deployment *might* happen with IPv6 because some of the issues have been solved (using, for example, embedable rendevous points - negating the need the 3rd parties). However if you look at how ISPs are architectured with xDSL networks, there isn't any incentive to provide multicast at the tail end.