scp vs ftp performance on SGI IRIX????????

Rick Jones rick.jones2 at
Fri Nov 7 06:27:40 EST 2003

Jay Chen wrote:
> Hi,
> Running OpenSSH 3.5p1 or 3.7.1p2 on two SGI orgin 300 systems of IRIX
> 6.5.17 with
> 4 600MHz CPU and gigabits network.
> Transfer a 1GB file with ftp took 28 seconds.
> Transfer a 1GB file with scp took 148 seconds.
> Use "snoop" to capture all network packets between these two systems
> . Ftp tansmitted about 107840 packets, but scp transmitted about
> 763339 packets.

So, 1GB is 1048576 KB.  1048576 KB divided by 107840 packets is 9956.8 bytes per
packet.  That seems rather odd because it is larger even than the bytes per
packet I would expect if JumboFrame (9000 byte MTU) were enabled.

Does Irix 6.5.17 do "large send" over the GbE NIC? Perhaps there is a
"sendfile()-like" call in Irix 6.5.17 that the Irix FTP uses

With the scp, that is more like 1406 bytes per packet, which is more consistent
with an MTU of 1500 bytes.  Or perhaps scp has some application-level block size
- I can never remember the specifics of SSL !-(

It is somewhat "consistent" that the case that sent ~7x the packets took ~7x the

> Also, did the same test on two IBM AIX 4.3.3 systems with 1 CPU and
> 10Mb network.
> Transfer a 1GB file with ftp took 1080 seconds
> Transfer a 1GB file with scp using blowfish took 1059 seconds.
> Ftp tansmitted about 207070 packets, but scp only transmitted only
> about 190290 packets.

In this case, it is showing >> 1460 bytes per packet which is I believe rather
impossible for a 10Mb network - I don't think any 10 Mbit/s NICs do jumboframes
and I'm confident that AIX 4.3.3 doesnot support large send - I believe that
came-in with patches to AIX 5.mumble (it is one of the reasons IBM has the
SPECweb scores they do)

Now, if there is compression of some sort I suppose that might explain some of
it.  Any chance that either ftp or scp were compressing the data?

It might be a good idea to take snapshots of the netstat statistics from before
and after each transfer - it is possible that the snoop traces aren't getting
all the packets.

You can then run the before and after netstats through "beforeafter" which you
can find at  source is included and
I suspect it would compile without too much trouble on Irix and AIX.

rick jones
Wisdom Teeth are impacted, people are affected by the effects of events.
these opinions are mine, all mine; HP might not want them anyway... :)
feel free to post, OR email to raj in  but NOT BOTH...

More information about the openssh-unix-dev mailing list