home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
kermit.columbia.edu
/
kermit.columbia.edu.tar
/
kermit.columbia.edu
/
newsgroups
/
misc.20020314-20021006
/
000002_fdc@columbia.edu_Thu Mar 14 13:31:47 EST 2002.msg
< prev
next >
Wrap
Text File
|
2002-10-06
|
5KB
|
101 lines
Article: 13267 of comp.protocols.kermit.misc
Path: newsmaster.cc.columbia.edu!news.columbia.edu!news-not-for-mail
From: fdc@columbia.edu (Frank da Cruz)
Newsgroups: comp.protocols.kermit.misc
Subject: Re: Experience with K95 over satellite links
Date: 14 Mar 2002 13:31:02 -0500
Organization: Columbia University
Lines: 84
Message-ID: <a6qq96$efo$1@watsol.cc.columbia.edu>
References: <c4914b87.0203140931.41ac24bc@posting.google.com>
NNTP-Posting-Host: watsol.cc.columbia.edu
X-Trace: newsmaster.cc.columbia.edu 1016130663 10236 128.59.39.139 (14 Mar 2002 18:31:03 GMT)
X-Complaints-To: postmaster@columbia.edu
NNTP-Posting-Date: 14 Mar 2002 18:31:03 GMT
Xref: newsmaster.cc.columbia.edu comp.protocols.kermit.misc:13267
In article <c4914b87.0203140931.41ac24bc@posting.google.com>,
satprof <satprof@altavista.com> wrote:
: Following my posting at the end of Jan, 2002 about evaluating Kermit
: for a project to transfer ~70MB of data nightly via a double(!)
: satellite hop, it might be useful for others in the future to hear of
: my experiences.
: ...
: The approach I adopted was to set up the K95D.EXE 'daemon' on the
: machine holding the data & to script the file transfer on the machine
: to which the data was to be sent. Before leaving for Africa, I tested
: the script (based on the 'reliable file transfer' script in the K95
: 'scripts' folder) between my home in France & my office in Geneva.
: With an ADSL link at home (& T3 at the office) my tests showed around
: 20KB/second. (Ping time ~ 40mSec)
:
: However, on arrival in Africa, I couldn't achieve more than about
: 3KB/second in practice. The receiving end showed only that a single
: window was in use, but that is apparently normal at the receiver.
:
Right. The reciever's window size goes up only when a packet is missing.
When the missing packet eventually shows up, it can write out its window
to disk and the window size goes back to one.
: The
: sending end showed that sliding windows were in use, but only 5 out of
: 30 possible. (Packet size settled down to 4000/3999; streaming was off
: to ensure block checking & the Fast macro was invoked.)
:
That means the sender was receiving acknowledgements within five packet
times. The window never grows larger than it needs to, and the sender
is pushing out packets as fast as it can. My guess is that your modem
was flow-controlling the PC, which would happen if the PAD was
flow-controlling the answering modem, which is not a stretch. Remember,
the PAD is taking each 4K packet and segmenting it into 32 128-byte
packets and sending them over an X.25 network with a window size of 7,
using go-back-to-n retransmission strategy rather than selective repeat
like Kermit. If the satellite links are dirty or lossy, this can result
in awful (but still presumably error-free) end-to-end throughput.
If you use hardware flow control, the modem lights would tell the tale.
If you weren't using modems at all, the same principals would still apply.
: The relatively slow speed was a bit of a disappointment, but it was
: overcome by initiating 5 concurrent instances of Kermit, and
: transferring the data in parts.
:
That's just the right thing to do in a situation like this. Each
connection is its own virtual circuit within the X.25 cloud and so does
not affect the others.
: The prime reason for the post is to record this for others, but I'd
: also like to thank Frank & the others in the group for their input at
: the beginning of the project.
:
Our pleasure :-)
: If anyone has any ideas about how to increase the no. of windows
: actually used, I'd like to try that out.
:
See above. Modern Kermits have a default maximum window size of 30, and
they use all the slots they need to use at any given moment. If the
number of slots used is low and your throughput is also low, it means
Kermit is being throttled by the communication medium.
If the Kermit-to-Kermit connection was over TCP/IP as I assume it must
have been since K95D.EXE is involved, you might have done better to
simply allow streaming. On the other hand you might also have done
better by making a directly dialed connection so as not to have the
differing transport strategies of Kermit, TCP/IP, and X.25 all fighting
each other at once.
: I also tried to get K95D.EXE
: to run as a WINNT service, but without success, so would be interested
: to know if anyone has managed this elsewhere.
:
This is quite straightforward in the new release, which is in its
final days of Beta testing:
http://www.columbia.edu/kermit/k95next.html
Thanks for the terrific report -- I wish everybody would do that!
- Frank