SCEF0003@WSUVM1.CSC.WSU.EDU ("James N. Petersen") (04/05/91)
In a recent article, Joe Strong <cca.ucsf.edu!jst@cgl.ucsf.edu> writes: >Would someone mind relating to me what kind of performance they've >gotten out of the KA9Q software on Ethernet, doing FTP, versus some >of the official commercial packages? >Machine1: DOS, 386/25, WD8003e, KA9Q software >Machine2: Xenix, 386/20, Excelan Board, Excelan TCP/IP package. >FTP put from 1 to 2 = 70KB/sec. >FTP put from 2 to 1 = 10KB/sec. >Is there some known bug, like WD8003E's not being able to swallow >back-to-back packets, that I'm missing here? Although I am using CUTCP and not KA9Q, I see the same sort of differences when using CUTCP and transfering to either an IBM 3090 or a DecStation 5000. Transfers from the DOS machine occur rapidly, transfers to it are much slower. I have assumed that it is because of the disk access speed on the PC, but perhaps I am wrong. Suggestions. Jim
mcdonald@aries.scs.uiuc.edu (Doug McDonald) (04/05/91)
In article <9104041437.aa27983@louie.udel.edu> SCEF0003@WSUVM1.CSC.WSU.EDU ("James N. Petersen") writes: >In a recent article, Joe Strong <cca.ucsf.edu!jst@cgl.ucsf.edu> writes: > >>Would someone mind relating to me what kind of performance they've >>gotten out of the KA9Q software on Ethernet, doing FTP, versus some >>of the official commercial packages? > >>Machine1: DOS, 386/25, WD8003e, KA9Q software >>Machine2: Xenix, 386/20, Excelan Board, Excelan TCP/IP package. > >>FTP put from 1 to 2 = 70KB/sec. > >>FTP put from 2 to 1 = 10KB/sec. > >>Is there some known bug, like WD8003E's not being able to swallow >>back-to-back packets, that I'm missing here? > >Although I am using CUTCP and not KA9Q, I see the same sort of differences >when using CUTCP and transfering to either an IBM 3090 or a DecStation 5000. >Transfers from the DOS machine occur rapidly, transfers to it are much >slower. I have assumed that it is because of the disk access speed on the >PC, but perhaps I am wrong. Suggestions. > >Jim An interesting comparison. I have noticed some slow transfers using NCSA Telnet involving our Vax. The rate was 1 kilobyte/sec. But they have changed their software. So I tired it today and got the following results: The PC is a Dell 310 20 MHz 386 clone with 8 megs memory and a 16 msec IDE disk, using Hyperdisk cache. Ethernet card is a WD 8003E. The Vax is some sort of recent miaco-Vax (3500????) running the current VMS. The MIPS is a MIPS 120. I am using NCSA Ftpbin on the PC here and transferring a 675 kilobyte file. send to VAX: 71 kilobytes/sec get from VAX to a file: 53 kilobytes/sec get from VAX to nul: 64 kilobytes/sec send to MIPS: 97 kilobytes/sec get from MIPS to a file: 107 kilobytes/sec get from MIPS to nul: 128 kilobytes/sec These are respectable results. If I change the PC from Hyperdisk to Smartdrive and run NCSA Telnet inside a window inside Microsoft Windows on the PC and then ftp BACK to the PC from the VMS or Unix machine I get: send to VAX: 61 kilobytes/sec get from VAX to a file: 29 kilobytes/sec get from VAX to nul: 52 kilobytes/sec send to MIPS: 66 kilobytes/sec get from MIPS to a file: 42 kilobytes/sec get from MIPS to nul: 73 kilobytes/sec These still seem reasonable. You will note that the VAX is as bad a bottleneck as Microsoft Windows!!!!! This has proven true even when ftping between the VAX and the MIPS. AS you might expect, I tend to avoid the VAX. Doug McDonald
jbvb@FTP.COM ("James B. Van Bokkelen") (04/06/91)
>>Machine1: DOS, 386/25, WD8003e, KA9Q software >>Machine2: Xenix, 386/20, Excelan Board, Excelan TCP/IP package. >>FTP put from 1 to 2 = 70KB/sec. >>FTP put from 2 to 1 = 10KB/sec. The PC is a Dell 310 20 MHz 386 clone with 8 megs memory and a 16 msec IDE disk, using Hyperdisk cache. Ethernet card is a WD 8003E. The Vax is some sort of recent miaco-Vax (3500????) running the current VMS. The MIPS is a MIPS 120. ... using NCSA Ftpbin on the PC here and transferring a 675 kilobyte file. send to VAX: 71 kilobytes/sec get from VAX to a file: 53 kilobytes/sec get from VAX to nul: 64 kilobytes/sec send to MIPS: 97 kilobytes/sec get from MIPS to a file: 107 kilobytes/sec get from MIPS to nul: 128 kilobytes/sec With PC/TCP 2.05, running on a WD8003/A in a 16Mhz Model 80, with the default number of packet buffers (5) and a 4Kb TCP window, talking to a Sun 386i on the same Ethernet, I can FTP files from disk to null device in either direction at about at 205Kb/sec. The more the disk matters in a benchmark of this sort, the less reliable it is; I don't know how fragmented either end of a reported transfer is... James B. VanBokkelen 26 Princess St., Wakefield, MA 01880 FTP Software Inc. voice: (617) 246-0900 fax: (617) 246-0901