> I'm not sure I totally agree. > > On Tue, Jun 24, 2008 at 2:07 PM, Jim Davis <[EMAIL PROTECTED] > com> wrote: > > USENet is actually pretty bandwidth inefficient beast. "Your" > server is peered > > with any number of other servers. Periodically your server asks all > its peers > > "got anything new?" - if so they send everything. > > The number of peers for a news farm is pretty small. The number of > peers for a P2P network is comparatively very large. My > understanding > is that most news goes through a small number of central hubs > now-a-days and it's very unlikely that TWC (for example) is peering > with 50 other nodes, but probably only 2-3 news major provider hubs. > These are also likely the same hub companies their users are going to > start using once they no longer have a local news source, effectively > converting traffic that was local to the ISP network into public > internet traffic. > > > In specific however each peer may choose to accept "everything", > only some > > groups or only certain messages (just new ones, or ones younger than > a > > certain age for example). In the end however if you've got a > network of a > > million peers then you've got a million copies of the message. > > There are almost certainly not a million news peers on the internet. > Nowhere close. > > > Since the servers have vastly different speed and storage profiles > you get huge > > diffences in "retention" - a huge server might be able to store a > year's worth of > > data while an average server only a month and a tiny server only a > day or two. > > True. > > > Compared to peer-to-peer which leaves the actual data on the "home" > servers > > until requested and, until then, just sends metadata USENet isn't as > resource > > friendly. Even a large binary message is shared amongst ALL peers > even if > > NOBODY ever requests it - if it's posted, it's transfered (again and > again and again). > > True, and these news propogation connections are (in theory) between > two ISPs on a very high bandwidth connenction. Home based data > transfers also go over this high speed connection, but at a the speed > of the slowest user's home connection, usually the upspeed max of > that > user divided by the number of upload fragments, which is very, very, > very slow in comparison. > > > USENet was also not designed for binary transfers (although they > make up > > the vast bulk of the bandwidth used) and need to convert any binary > files to > > text - this often expands the files making them bigger in > transmission than > > in the native format. > > True. > > > Because of the uncertainty of USEnet (retention rate, peerage, etc) > most groups > > have adopted procedures for ensuring quality transfers. One of > these is provide > > PAR (parity) files for large binaries. Using these files any > corrupted or missing > > message parts can be rebuilt (as far as I'm concerned it's > witchcraft pure and > > simple). The trouble is a full "PAR set" needs to be, usually, > about 20% of the > > file size. It's also generally accepted practice to include a > sample of any media files. > > True. > > > So a 700 Meg movie downloaded from peer-to-peer would be 700 meg. A > 700 Meg > > movie on USENet is 750 meg converted to text, plus, say 160 meg of > PAR files > > plus maybe a 30 meg sample file. And, again - these are all copied > to every peer > > regardless of demand. A 700 Meg movie on a peer-to-peer network > that nobody > > ever requests uses essentially zero bandwidth. > > Okay, let's accept those numbers and do some math. Given 1000 > clients > downloading the same 700 meg file two ways. Either P2P clients or > via > News. Let's say there are 100 news servers, each feeding 10 people. > > 1) First, P2P - 1000 clients on a 700mb file at a 1:1 UL/DL ratio > - 700,000mb over the public internet > - 700,000mb over local ISP network > > 2) Next Usenet - 100 news servers transferring 780 > (file+encoding+pars) a max of 99 times > - 77,200mb over the public internet > - 780,000mb over local ISP lines > > 3) Last, consider my numbers are wrong and it's 10 news servers > feeding 100 people each > - 7,020mb on public internet > - 780,000mb over local ISP lines > > Even the worst usenet case (2) leaves 622,800 spare mb of internet > bandwidth for those messages that get downloaded and never read off a > server. That's 89%. If 89% of the messages on a news server were > never read, P2P would only just break even with usenet. > > Yes, the local bandwidth is higher, but the trade off is worth it > methinks. > > > USENet is also limited geographically: you connect to whatever > server you > > connect to. A closer server (even a peer) might be faster, but you > don't have > > access. > > Only if you are using a public server and not a local ISP server. > > > Peer-to-peer attempts to optimize the connection by downloading > files from > > the closest/fastest peer that hosts them and downloading using > multiple hosts. > > Once you have the file you also become a new node, a new host, for > that file > > thus allowing the system to optimize even further. Once I download > something > > from USENet it's mine and nobody else can see it. > > True, but all that P2P traffic is still public internet traffic. > Usenet traffic is primarily local ISP traffic. > > > In short USENet servers act a lot like a peer-to-peer network where > everybody > > is constantly requesting every file from everybody else. For the > end user > > USENet is more like the Web - a provider and a consumer with a very > large > > downstream and very small upstream data flows. > > I'd say more like Akami than the general web. Again, news servers > don't peer with "everyone", just certain peers they have agreements > with. At the ISP I worked at just after college, we actually only > peered with one provider. > > > This is simplistic of course - there are all sorts of tricks. > On-the-wire > > compression helps text-only USEnet more than peer-to-peer and much > > of USENet traffic is still run over the relatively open backbone > connections - > > it's the last mile connections that are truly getting clogged and > these are > > exactly what's leveraged most by peer-to-peer. > > Yup. Plus, usenet doesn't saturate your upstream like P2P > > > But in the end I'm still a little suprised that USENet doesn't have > a stronger showing. > > Heh - longer reply than I meant, but I am still not surprised. ;) > -Cameron
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~| Adobe® ColdFusion® 8 software 8 is the most important and dramatic release to date Get the Free Trial http://ad.doubleclick.net/clk;203748912;27390454;j Archive: http://www.houseoffusion.com/groups/CF-Community/message.cfm/messageid:262637 Subscription: http://www.houseoffusion.com/groups/CF-Community/subscribe.cfm Unsubscribe: http://www.houseoffusion.com/cf_lists/unsubscribe.cfm?user=11502.10531.5
