home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!destroyer!gatech!usenet.ins.cwru.edu!odin!trier
- From: trier@odin.ins.cwru.edu (Stephen C. Trier)
- Newsgroups: comp.protocols.tcp-ip.ibmpc
- Subject: Re: Fragmented IP packets: any PD implementations?
- Date: 8 Nov 1992 21:25:39 GMT
- Organization: Case Western Reserve University, Cleveland, OH (USA)
- Lines: 38
- Message-ID: <1dk0kjINN410@usenet.INS.CWRU.Edu>
- References: <1992Nov8.203621.6902@mnemosyne.cs.du.edu>
- NNTP-Posting-Host: odin.ins.cwru.edu
- Keywords: IP fragments TCP/IP
-
- In article <1992Nov8.203621.6902@mnemosyne.cs.du.edu> dpayne@nyx.cs.du.edu (Dirk Payne) writes:
- >In order to be a "real" TCP/IP package (connected to the "real" world)
- >fragments cannot be ejected out of hand.
-
- In the real world, fragments are not terribly important. TCP intentionally
- avoids fragmenting data. The only protocol I know of that intentionally
- sends fragments is NFS, which is supported by none of the packages you named.
-
- I am writing this using a PC/IP derivative that does not support fragments.
- We have had 3000 users running it for two years, and in that time, I have
- had only one complaint about a connection that failed because fragments were
- being sent.
-
- >Has anyone implemented IP fragment handling in any of the above mentioned
- >products?
-
- As far as I know, the only "free" TCP/IP package that supports fragment
- reassembly is ka9q. (It's not free to everyone -- check the docs for
- details.)
-
- Fragment reassembly is missing from DOS IPs for two reasons. First, it
- adds substantial complexity for little payoff, as exemplified by how well
- these packages function without doing fragments. Second, reassembly isn't
- useful unless it can handle packets several times the MTU. Memory is
- tight on MS-DOS systems, so most TCP/IP implementations go to great lengths
- to conserve RAM. Reassembly requires being able to allocate relatively
- large amounts of RAM, RAM that is wasted if no fragments arrive.
-
- Someday, our PC/IP derivative may support reassembly. I would expect that
- it is on the wishlists of the other TCP/IP authors, too. It's just not
- easy to do, and it is hard to justify the time and memory for reassembly
- when one can satisfy the vast majority of one's users without it.
-
- --
- Stephen Trier
- Network Services Engineering, IRIS/INS/Telecom "Dessine-moi un mutton"
- Case Western Reserve University - Le Prince
- trier@ins.cwru.edu
-