Christian Huitema wrote:

> The first SYN packet gets lost, and
> the client simply picks another address in the list and tries again.

The APIs I've used don't tell me about lost SYN packets (thank goodness); they
only tell me if the connection has timed out.

> So, yes, we have a problem. We need to somehow adapt the TCP stack to
> survive losses of transit networks. But we should not overstate that
> problem. It only affects long connections,

(a) But long connections are important; they're more efficient (and give the
users better performance) than short connections.  Modern application
protocols are being designed with this in mind.

(b) It also affects short connections, just not as often.  If Joe User's HTTP
connection gets dropped, he'll see "Transfer interrupted" and think there's
something wrong with the server (or he'll see a broken image and think the
site's HTML is messed up).  It would not occur to him that the trouble would
be in a transit network; he's probably never heard the term.  So he'll come
away thinking that the Internet is just plain flaky.

> it only makes a difference if a
> connection to a transit provider breaks,

Or if the chosen path becomes congested over time.

> In any cases, there are simple modification to
> TCP, for which we already have some experience, that could handle the
> problem. In the long run, once these modifications are in place, we are
> better off than in the current situation,

OK.  How long is the long run? How long did it take to get the LFN fixes
deployed? They were described in 1989 (RFC-1106); I seem to remember they
weren't widely available as of 1994 or so.  (My memory may be skewed, though,
because one of the machines I was using was running SunOS 4.x, which wasn't
being updated much.)

--
/==============================================================\
|John Stracke    | http://www.ecal.com |My opinions are my own.|
|Chief Scientist |=============================================|
|eCal Corp.      |They prayed for their fates to be quick,     |
|[EMAIL PROTECTED]|painless, and, ideally, someone else's.      |
\==============================================================/


Reply via email to