home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!ogicse!das-news.harvard.edu!cantaloupe.srv.cs.cmu.edu!crabapple.srv.cs.cmu.edu!andrew.cmu.edu!tm8t+
- From: tm8t+@andrew.cmu.edu (Tod McQuillin)
- Newsgroups: comp.protocols.tcp-ip
- Subject: Re: Transfer of 100's of MBytes files
- Message-ID: <8eyndsC00XQDELNoRM@andrew.cmu.edu>
- Date: 6 Nov 92 17:39:36 GMT
- Article-I.D.: andrew.8eyndsC00XQDELNoRM
- References: <rjberke.38.0@berke.itg.ti.com>
- Organization: Carnegie Mellon, Pittsburgh, PA
- Lines: 46
- In-Reply-To: <rjberke.38.0@berke.itg.ti.com>
-
- rjberke@berke.itg.ti.com (Richard Berke) writes:
- > How often (1 out of 20?) do you run into aborted transfers that
- > have to be restarted from scratch?
-
- Pretty often. Especially with large files transferred over a SLIP
- link. I'd say 1 out of 50 times over SLIP, and 1 out of 200 on more
- permanent connections.
-
- > Have you been able to distinguish
- > what portion of those failures are due to disk issues, circuit failures,
- > or protocol implementations?
-
- In my experience, TCP is an extremely reliable protocol. Aborted
- transfers are almost always caused by network hardware problems, or
- routing difficulties.
-
- > I am frequently asked about the reliability of using FTP to move
- > files and/or sets of files that are 200-400 Mbytes, and there's interest
- > in some means of recovering from a disruption in the middle of a transfer.
-
- You can always check that the transfer was successful by taking the
- checksum of the data on both ends. The more recent implementations of
- the FTP protocol I've seen all support the "reget" command (I believe
- the FTP spec calls this "RETR") which allows you to resume an
- interrupted transfer. I use it all the time and have never had a
- problem with it.
-
- > I'm sensitive to disruptions, but wary of exotic schemes to recover.
-
- I'd hardly call the reget command exotic.
-
- > Since our world-wide links are as small as 56 Kbits/sec, transmission times
- > are significant. Good throughput for FTP (not best case) is about 5 Kbytes
- > per second. 200-400 Mbytes takes 1-3 hours or more. Detection of failed
- > transfers may be delayed, further adding to wall clock time taken to
- > accomplish the overall transfers.
-
- You have a point there. The reget command does not automatically work
- -- you have to type it yourself.
-
- > I've heard that protocols exist for transfers that use checkpoints and
- > can restart. I've never heard of mainstream suppliers for them.
- > Are they available? How necessary are they, compared to FTP?
-
- I've never had a need for something like this, but then, it sounds
- like your data-moving requirements are more demanding than mine.
-