home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.protocols.tcp-ip
- Path: sparky!uunet!spool.mu.edu!news.cs.indiana.edu!lynx.unm.edu!nmsu.edu!amolitor
- From: amolitor@moink.nmsu.edu (Andrew Molitor)
- Subject: TCP congestion/Nagle's algorithm
- Message-ID: <1993Jan23.113544.5652@nmsu.edu>
- Originator: amolitor@moink.nmsu.edu
- Sender: usenet@nmsu.edu
- Organization: Department of Mathematics
- Date: Sat, 23 Jan 1993 11:35:44 GMT
- Lines: 25
-
-
- Darn it, I think I put my foot in my mouth. After poking
- through rfc896 (in which Nagle introduces his algorithm for
- solving the 'small packet' problem) and comparing with Karn's
- NOS TCP implementation, I find myself confused.
-
- First, I misunderstood congestion windows. My apologies
- to whoever it was I responded to a day or so ago, blithely
- dismissing throughput problems as 'sounds like a small congestion
- window.'
-
- Second, Nagle seems to be saying that a TCP should not
- send new data while unack'd data is outstanding, while Karn's
- code seems to not send new data while unack'd data is outstanding
- *unless* there is a max segment sized chunk to go out.
-
- Karn's version seems right, since it allows proper
- streaming, while Nagle's seems to reduce the sliding window
- to a stop and wait arrangement. Is Karn's algorithm in Nagle's
- remarks, disguised as a tacit assumption that one always sends
- all the data one has, in multiple segments? He seems to be implying
- that as soon as one gets an ack, one sends all the data one has
- on hand. This is still not quite Karn's implementation, I think.
-
- Andrew
-