I'm really suprised that no-one actually read what I wrote. Let me make it really simple, as it seems no-one can read something until the end anymore (I presume no-one gets to do the test paper where the last "question" says "on this occasion to pass you must not complete any of the above questions").
Bullet points, people: * proxy servers - bad * store and forward - good * Doing redirect using DNS - bad * File synchronization - great, fast in backbone network * Differential file sync - 100% unnecessary So: * ISPs provide rack space for BBC servers inside their network * ISPs provide list of IP addresses to directed to said servers * BBC copies each new file (and deletes) to these servers * iPlayer software detect and redirects to BBC servers inside ISP network * Interim solution until fatter pipes purchased, say 2-3 years. On 15/04/2008, Darren Stephens <[EMAIL PROTECTED]> wrote: > > Just a few thoughts (some of which may be emanating from my posterior, > but no matter): > > > > *From:* [EMAIL PROTECTED] [mailto: > [EMAIL PROTECTED] *On Behalf Of *Brian Butterworth > *Sent:* Monday, April 14, 2008 5:38 PM > *To:* [email protected] > *Subject:* Re: [backstage] iPlayer and the ISPs - a solution > > > > > > but my experience of them is that transparent proxies reduce overall > > performance because they need to get in the way of each and every HTTP > > transaction. > > Yes, I suppose in theory, but use of appropriate routing and firewalling > means not in practice. Though adding this before the application layers will > introduce a separate latency of its own. It would be difficult to filter > such traffic on port admittedly, but not on other things, like source and > dest addresses (see below). > > > I wouldn't have thought that the small increase in latency would be > noticeable for a several hundred megabyte file. > > I would have thought otherwise, since the latency is, almost by > definition, indeterminate and could, in fact, be appreciable, especially if > under high load > > > > 3. Store and forward: Locate MIRROR SERVERS inside the ISP network. > > This seems a much better idea. > > But the BBC's network does a LOT of this mirroring and load balancing > stuff already, certainly if you look at some parts of their operation (like > News) and especially with HTTP. > > It wouldn't work otherwise. And when it doesn't quite work like that, > performance does suffer. > > > It sounds a lot like some kind of Cache. And another question is *who* > is going to pay for the servers that speak RTMP? This sounds like some > kind of revenue driving scheme for the BBC's commercial friends. > > > > the ISP provide the BBC with rack > > space 'inside' their networks for mirror servers. > > > > A generic cache would be much more scalable, if the servers only mirror > BBC data then this does nothing to solve problems with other sites. > > How does one mirror this data? Will it be available via rsync? Will it > be mirrorable by *anyone* or does the BBC intend to pick and chose > commercial ISPs to provide better access to. Again very shaky ground. > > And even though technologies like rsync are largely differential, the > traffic generated from such syncing is not trivial, especially if the > content is in binary formats and not textual. Because constructed deltas > that are used for syncing may not be that small. And, more prosaically, once > the data is inside the ISP's networks, who is responsible for it? > > > > - change the main BBC iPlayer to redirect requests for the content to > > the Mirror Server located in the ISPs network. > > Really unscalable, how is the BBC going to know which ISPs have mirrors > and which do not? This would require each ISP to notify the BBC. Just > seems wrong. Having every Content Provider have to speak to every ISP > seems to go against the core of the Internet. > > > > If the BBC is decides to provide such a service, what is wrong with it > whitelisting those who sign up to use it? > > Not necessarily something I agree with but not unfeasible from a technical > point of view. Potentially very fiddly however and, as rightly pointed out, > not hugely scalable for the long term. > > > If a pipe on the Internet is not running at 100% it is being underused! > > > > On the other hand, a pipe running at 100% could clearly be considered > borderline congested. > > > > Andy > > [1] > < > http://www.opsi.gov.uk/acts/acts1998/ukpga_19980041_en_2#pt1-ch2-pb2-l1g18 > > > > - > Sent via the backstage.bbc.co.uk discussion group. To unsubscribe, please > visit http://backstage.bbc.co.uk/archives/2005/01/mailing_list.html. > Unofficial > list archive: http://www.mail-archive.com/[email protected]/ > > > > > -- > Please email me back if you need any more help. > > Brian Butterworth > http://www.ukfree.tv > > > ***************************************************************************************** > To view the terms under which this email is distributed, please go to > http://www.hull.ac.uk/legal/email_disclaimer.html > > ***************************************************************************************** > -- Please email me back if you need any more help. Brian Butterworth http://www.ukfree.tv
