On Tue, Jan 23, 2001 at 12:12:12PM +0100, P Lanvin wrote:
> Yes, this build is a gcc. It _really_ should not be a compiler issue.
> Hopefully the trusslogs will say something useful.
I've been messing with this issue for the past couple of weeks. I'm
running tests on two Ultra5s on switched 100mbit ethernet, with cipher
specified as blowfish and no compression. Using scp1 (1.2.27 from
ssh.com), i get a little better than 1 Mb/s in either direction.
Fine.
Using scp2 (2.3.0 from ssh.com), if i'm doing a "get-style" transfer
(as in, copy remote file to local directory), i get over 1 Mb/s. But
if i'm doing a "put-style" transfer, i get 300 Kb/s max. in other words:
% scp -c "blowfish" remote:remotefile localdir/
is okay, while
% scp -c "blowfish" localfile remote:remotedir/
This makes no sense to me.
- Snooping the line indicates that substantially more packets are being
sent in the "put" case, though not a substantially larger amount of
data.
- Running the command with debugging cranked up indicates that almost
twice as many lines are being logged in the "put" case. That
doesn't, of course, mean much about the actual amount of work that's
being done, and unfortunately don't have time or expertise to wade
through the code and find out what the deal is.
- Yes, i get comparable results if i switch which machine i'm running
the commands from. So it's not just a case of one machine being
- i get the same results if i do "cat file | ssh remote "cat >file", so
it's not a question of the scp wrapper to ssh slowing things down.
Does anyone who's looked in the code know anything about this? Can
anyone else replicate it? Help!
-Chaos
--
------------------------------------------------------------------------
E. Chaos Golubitsky ([EMAIL PROTECTED])
System Administrator
Harvard-Smithsonian Center for Astrophysics
Phone: (617)-495-7144
