NetNews Usenet Archive 1993 #1/NN_1993_1.iso/spool/comp/arch/12201

Internet Message Format | 1993-01-09 | 5.6 KB

Xref: sparky comp.arch:12201 comp.benchmarks:1922
Path: sparky!uunet!zaphod.mps.ohio-state.edu!magnus.acs.ohio-state.edu!usenet.ins.cwru.edu!agate!doc.ic.ac.uk!uknet!ieunet!tcdcs!unix1.tcd.ie!bmccnnll
From: [e-mail] (Barry McConnell)
Newsgroups: comp.arch,comp.benchmarks
Subject: Re: FTPable Performance Results
Message-ID: <[e-mail]>
Date: 10 Jan 93 02:02:38 GMT
References: <[e-mail]>
Sender: [e-mail] (NN required at ashe.cs.tcd.ie)
Organization: Trinity College, Dublin
Lines: 100
Nntp-Posting-Host: unix1.tcd.ie

In <[e-mail]> [e-mail] (Vincent Cate) writes:

[...]

>What other FTPable and current benchmark papers are out there?

Here's something I had lying around...

From: [e-mail] (Skip Sauls)
Newsgroups: comp.sys.amiga.advocacy
Subject: New Network "Benchmarks"
Date: 1 Feb 92 08:25:28 GMT

These are the results of a series of benchmarks to determine the speed at which
various computers can transfer files via FTP over Ethernet.  Previously, I
posted a similar set of figures which were hastily generated.  This set took
quite a while and this is a big post, so hit "n" if you are not interested.


If you decided to stick around, I hope that this is not too boring.  I am
studying TCP/IP this semester and finally learning how all of this wonderful
network stuff really works.  These tests were conducted on a Friday evening
with no one else around to tie up the network, so the results should be valid.

Basically, I used the FTP software on several machines to transfer a file back
and forth with "get" and "put" using binary mode.  The file is 375048 bytes,
and if you are a typical Amiga owner, you probably have this exact same file
somewhere and may use the application every day.  This file is larger than the
largest buffer on the Ethernet cards on all computers tested.  By using a file
which is ~25k, the transfer rate improves somewhat, but a file that is ~750k is
transferred at the same rate as the test file.



Here are the players:

Code        A1         S1         S2         M1         P1         P2
Make        Commodore  Sun        Sun        Clone      Clone      NCR
Model       A3000      4/110      3/60       486 Tower  386        PC 386
OS          ADOS 2.02  SunOS 4.1  SunOS 4.1  Mach386    MSDOS 5.0  MSDOS 3.3
CPU         68030      Sparc      68020      80486      80386      80386
FPU         68882      Sparc      none       80486      none       80387
MHz         25         14.32      16         33         25         16
RAM         4M/2M      32M        16M        16M        4M         6M
Bus Type    Zorro III  VME        none       ISA        ISA        ISA
HD Type     SCSI       SCSI       SCSI       SCSI       IDE        ST506
HD Size     52M        1G         150M       660M       89M        30M
ENet Make   Commodore  Sun        Sun        3Com       3Com       3Com
ENet Size   16 bit     16 bit     16 bit     16 bit     16 bit     16 bit
Video       ECS        Sun 8 bit  Sun Mono   SVGA       SVGA       EGA

All machines are on a local Thin Ethernet network.  The Suns and the 486 are
connected S1<->M1<->S2 with 2x5 feet of cable, the PCs are connected P1<->P2
with 5 feet of cable, and P2<->S1 with 30 feet of cable.  The Amiga is on the
other side of the building and is connected by more than 150 feet of cable.


The codes A1, S1, S2, M1, P1, and P2 are used to identify the computers on the
chart below.  The leftmost column and topmost row indicate the machine from
and to the getting and putting is done, respectively.  The figures show the
transfer rate in kbytes/second.

         A1        S1        M1        S2        P1        P2

A1 get   147       147       147       147        10         9
   put   147       118       147       118        14        25
S1 get   118       610       216       407         9        17
   put   147       610       262       407        17        21
M1 get   141       229       262       262         9         9
   put   131       282       229       282        15        14
S2 get   115       407       305       305         9        10
   put   147       458       282       305        17        21
P1 get     5         5         5         5         *         5
   put     9         9         9         9         *         8
P2 get     5         5         5         5         5         *
   put     9         9         9         9         9         *

* = Test Failed


There does not appear to be a single factor which is responsible for the wide
range of transfer rates.  I would not blame the slow speed of the PCs on the
ISA bus alone, because the 486 running Mach also has an ISA bus and it has very
good performance.  All of the machines except for the two PCs are using a SCSI
bus for hard drives, so that may be a big factor.  After using the PC with a
fast hard drive and IDE interface, I would recommend that people stay away from
IDE and ignore all of the hype surrounding it.  It is ironic that the 3 year
old 16MHz 386 with ST506 is faster than the 4 month old 25MHz 386 with IDE.
BTW, the 16MHz has an NCR motherboard and the 25MHz is a Micronics, so these
are not junk built in some garage.

I would be interested in similar sets of figures for other machines if anyone
has the time to conduct the tests.  If I can find the time, I will move the
Amiga into the same room as the Suns and 486 and see if the distance is a major
factor.  The net-guys around here claim that it is, and I would like to see
how well the Amiga does on equal footing.

Skip Sauls              //    Graduate Student and Instructor
[e-mail]     \\ //    Department of Computer Science
A3000 + Ethernet = Freedom     \X/    University of Mississippi