kwang@bud.UUCP (Kwang Sung) (11/22/88)
Recently, I improved "ftp" in one of our products which is 4.2BSD based.
When I typed ftp> bin
ftp> get /syst /dev/null,
I had:
351465 bytes received in 1 second ( 3.4 e + 02 Kbytes/sec )
I didn't use Van Jacobson's Algorithm.
Is there anybody has better "ftp" performance than mine ???
Bye Bye
Kwang Sung
Sr. Software Engineer
ARIX Corp
408-922-1822
408-251-5040(H)
UUCP: ...!sun!aeras!smaug!kwang
.van@HELIOS.EE.LBL.GOV (Van Jacobson) (11/23/88)
> Recently, I improved "ftp" in one of our products which is 4.2BSD based. > I had 351465 bytes received in 1 second ( 3.4 e + 02 Kbytes/sec ) > ... > Is there anybody has better "ftp" performance than mine ??? It would be interesting to see what performance you would get with a good tcp. I think Dave Borman of Cray Research holds the current ftp speed record: He routinely gets 30 Megabytes/sec (of course, some of us think using two Cray-IIs gives him an unfair competitive advantage). If you're using more conventional hardware, 340 KB/s still isn't spectacular. Here's a typescript between two of our Sun 3/60s across an ethernet. Both machines are running stock 4.3bsd-tahoe ftp & ftpd with Mike Karels' and my tcp & kernel hacks: Script started on Wed Nov 23 02:40:44 1988 [vs 1]% ftp yak Connected to yak.ee.lbl.gov. 220 yak FTP server (Version 4.27 Sat Nov 5 02:20:42 PST 1988) ready. Name (yak:van): 331 Password required for van. Password: 230 User van logged in. ftp> bin 200 Type set to I. ftp> get /usr/lib/libcore.a /dev/null 200 PORT command okay. 150 Opening data connection for /usr/lib/libcore.a (669336 bytes). 226 Transfer complete. 669336 bytes received in 0.82 seconds (8e+02 Kbytes/s) ftp> get /usr/lib/libcore.a /dev/null 200 PORT command okay. 150 Opening data connection for /usr/lib/libcore.a (669336 bytes). 226 Transfer complete. 669336 bytes received in 0.82 seconds (8e+02 Kbytes/s) ftp> bye 221 Goodbye. [vs 2]% exit script done on Wed Nov 23 02:41:23 1988
jam@RADC-LONEX.ARPA (11/23/88)
>From: aeras!bud!kwang@sun.com (Kwang Sung) > >Recently, I improved "ftp" in one of our products which is 4.2BSD based. >When I typed ftp> bin > ftp> get /syst /dev/null, >I had: > 351465 bytes received in 1 second ( 3.4 e + 02 Kbytes/sec ) > >I didn't use Van Jacobson's Algorithm. > >Is there anybody has better "ftp" performance than mine ??? Huh? Your going to have to provide a little more than this if you want people to sit up and take notice! First of all, what did you do (in general terms if not specifics) to "improve" the ftp on your 4.2 based system? Ftp is not all that complicated when it transfers a file. I find it hard to see much room for tremendous speed increase at the application level. Secondly, what kind of networking are you using? ethernet? other? Third, were you staying on your net, or going to an internet host? In order to compare anything (be it ftp or ?) you have to have a compatible situation. For instance, if I ftp to Europe, it ain't never gonna keep up with calling the guy next door. Please elaborate if you desire to continue this. Joel A. Mussman Contel Federal Systems (this company has no opinion on ftp)
SCOTTY@UOGUELPH.BITNET (Steve Howie) (11/23/88)
We got approx 160 Kbytes/sec using FTP under VM in loopback i.e. FTP'ing from a CMS userid to the FTP server. This is effectively a very fast file copy. Mind you, this doesn't go out through our 8232 Ethernet controller :-) Scotty
dab@opus.cray.com (Dave Borman) (11/23/88)
I can't resist. Kwang Sung, from ARIX Corp, writes: > Recently, I improved "ftp" in one of our products which is 4.2BSD based. > When I typed ftp> bin > ftp> get /syst /dev/null, > I had: > 351465 bytes received in 1 second ( 3.4 e + 02 Kbytes/sec ) > > I didn't use Van Jacobson's Algorithm. > > Is there anybody has better "ftp" performance than mine ??? I am assuming you are going through the software loopback driver. Well, on one of our machines (this one was also running 4.2 based networking code at the time) I got: ftp> bin 200 Type set to I. ftp> get xxx /dev/null 200 PORT command okay. 150 Opening data connection for xxx (127.0.0.1,1028) (17444448 bytes). 226 Transfer complete. 17444448 bytes received in 1.34 seconds (12673.608 Kbytes/s) The numbers are better with our newer 4.3BSD based code. So I guess the answer to your question is yes. -Dave Borman CRAY Research, Inc.