From: torbjorn.lindgren@none.invalid   
      
   Jorgen Grahn   wrote:   
   >On Thu, 2015-09-24, bit-naughty@hotmail.com wrote:   
   >> 1) Suppose I'm watching a Youtube video - it's true that different   
   >> bytes of the same video could travel via different routes to get to   
   >> me, right? Say, I'm in the UK, one byte could come straight from the   
   >> US, and another via South America, say....? I mean, that's what TCP/IP   
   >> does, right?   
   >   
   >IP is designed so it /could/ happen, yes.  Not on the byte level but   
   >on the IP packet level.  And an application or IP stack which cannot   
   >cope with it happening is broken by definition, IMO.   
   >   
   >I think it doesn't happen very often.  Especially spreading one data   
   >flow over several routes has drawbacks, like packets arriving in the   
   >wrong order, sudden MTU changes and so on.   
      
   Splitting packets from the same "conversation" over multiple links is   
   relatively rare (to avoid packet reordering and various other   
   undesirable effetcs), but multiple simultaneous streams between two   
   machines can sometimes take different or very different paths.   
      
   Sometimes existing connections will shift path "mid-stream", due to   
   router/links either being removed or added. It's very rare that is   
   even noticeable if it happens.   
      
   Multiple links showing through in traceroute can make it harder to   
   debug, mostly if the paths are really diverse or different lenghts   
   (a bad, bad idea!).   
      
      
      
   >Perhaps it's more common that packets A->B take a different path than   
   >B->A does?  I'd be interested to hear about it from someone who knows   
   >more about the topology as it looks today.   
      
   I'd suspect MOST not directly peered US-only traffic actually goes   
   different paths in each direction, either different physical paths,   
   via different ISPs or both.   
      
   If we take a sufficiently large US ISP and a very large content   
   network (like Youtube or Netflix) they're probably peering directly to   
   each other (to reduce cost) and "popular" content either gets   
   delivered from a nearby node or is served from specialized caches   
   actually on the ISPs network.   
      
   In both those cases it's fairly likely it's traversing similar or   
   possibly identical paths in both directions which was why I excepted   
   them from my initial statement.   
      
      
   I could write a lot more on this subject but the best reference I know   
   is the traceroute presentation shown on the 47th NANOG Meeting (North   
   American Network Operators Group).   
      
   There's a reason why it's a recurring topic on the NANOG mailing list   
   and meetings, asymmetrical and "funky" routing is common in the real   
   world but normal text books doesn't cover it much.   
      
   https://www.nanog.org/meetings/nanog47/presentations/Sunday/RAS_   
   raceroute_N47_Sun.pdf   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)