Contents | < Browse | Browse >

===========================================================================
                         Review: MIAMI vs.  AMITCP
  Robert Davis                                             rdavis@nyx.net
===========================================================================

Here is a comparison of the performance of MIAMI as compared to AMITCP.  In
July, 1996, Holger Kruse released the beta and then demo versions of his
new TCP/IP stack for the Amiga, named MIAMI.

The primary advantages to MIAMI over the older AMITCP are:
1. MIAMI is *much* easier to install, often taking
   fewer than 10 minutes from start to on-line.
2. MIAMI is less expensive to register, costing $35 (US)
   while the commercial version of AMITCP/IP is around
   $95 (US) from retail sources.

The tests I have run show disagreements with the prevailing wisdom as seen
in Usenet newsgroups and on Internet Relay Chat channels.

The buzz seemed to be that MIAMI was simply faster with its Internet
connections than was AmiTCP, and that MIAMI used less memory than AmiTCP.

Neither is true.

Anyone can do these tests, all it takes is a bit of effort and some time. 
I would like to see someone come up with different results, using a test
procedure which I (or anyone) could duplicate.

Still, differences are slight, and the two advantages for MIAMI listed
above are very important.  Having a choice in TCP/IP implementations is
also important.

Here are the results of my comparitive tests of MIAMI and AmiTCP, starting
with an examination of memory used.

Procedure followed: Reboot the Amiga 3000, take note of free memory, start
the TCP/IP stack, start IBrowse, close unnecessary windows.  Note free
memory, subtract from original free memory to determine actual usage.

Software versions used were MIAMI 1.0 demo and AmiTCP/IP 4.0 demo, and
IBrowse R8a demo.

IBrowse is loaded because it uses MUI, and forces similar memory
allocations in both MIAMI (which uses MUI at startup) and AMITCP (which
does not use MUI).

case 1   MIAMI + IBROWSE
    memory used   chip 226976  other 1696896  total 1,923,872

case 2   AMITCP + IBROWSE
    memory used   chip 216128  other 1544048  total 1,760,176

case 3   MIAMI + IBROWSE
    memory used   chip 226976  other 1697088  total 1,924,064

case 4   AMITCP + IBROWSE
    memory used   chip 216128  other 1544048  total 1,760,176

Conclusion, once MUI is forced, AMITCP uses about 163-thousand bytes less
than does MIAMI.

====================================================================

The results of three speed comparisons between MIAMI (the new Amiga TCP/IP
system for modem use) and AMITCP/IP

22 July 1996

Speed of FTP file transfer ...  AmiTCP loaded, amiftp started, connected to
ftp.amigalib.com, received file INDEX.Z 941966 bytes.  the reboot loaded
MIAMI, and receive the same file from the same ftp site.  Then repeat.

attempt 1   AmiTCP  1365 bytes/second
attempt 2   MIAMI   2616 bytes/second
attempt 3   AmiTCP  2898 bytes/second
attempt 4   MIAMI   2366 bytes/second
attempt 5   AmiTCP  2943 bytes/second
attempt 6   MIAMI (transfer stopped at 95232 bytes)
attempt 7   MIAMI   2660 bytes/second

Interesting.  The very slowest and the very fastest ftp transfers were
while using AmiTCP.

I suspect that conditions on the Internet have a lot more to do with data
transfer speed than does which TCP stack you are running.

Note that attempt 6, using MIAMI, errored and failed.

====================================================================
02 August 1996

Some results from limited comparisons between MIAMI (demo version 1.0) and
AmiTCP (4.0 demo)

Speed of FTP file transfer ...  reboot the A3000, Load MIAMI, start amiftp,
connect to ftp.amigalib.com, receive file INDEX.Z 953269 bytes.  Reboot the
A3000, load AMITCP, and receive the same file from the same ftp site.  Then
repeat.

attempt 1   MIAMI   2633 bytes/second
attempt 2   AmiTCP  2868 bytes/second
attempt 3   MIAMI   2862 bytes/second
attempt 4   AmiTCP (transfer aborted at 670720 bytes)
attempt 5   AmiTCP  2812 bytes/second
attempt 6   MIAMI   2862 bytes/second
attempt 7   AmiTCP  2597 bytes/second

Interesting.  For the second time, the very slowest and the very fastest
ftp transfers were while using AmiTCP.  But MIAMI is very close ...  I
conclude there is no significant difference in the speed of data transfer
using MIAMI or AmiTCP.

I still suspect that conditions on the Internet have a lot more to do with
data transfer speed than does which TCP stack you are running.

This time attempt 4, using AMITCP/IP, errored and eventually failed.

====================================================================
05 August 1996

Again using MIAMI 1.0 demo and AMITCP 4.0 demo, here are the results of a
third speed test.

Slightly changed conditions for the third speed test involve putting
INDEX.Z (954859 bytes) on the local Internet Service Provider.  That LINUX
system is about one mile from my home.

The same procedure, reboot, load a TCP/IP stack, receive the file, then do
it all again, gave the following results:

attempt 1  MIAMI  3281 bytes/second
attempt 2  AMITCP 3304 bytes/second
attempt 3  MIAMI  3236 bytes/second
attempt 4  AMITCP 3304 bytes/second
attempt 5  MIAMI  3304 bytes/second
attempt 6  AMITCP 3304 bytes/second

Certainly, neither MIAMI nor AMITCP/IP has a significant speed advantage. 
Speed variations here might be attributed to load differences on the remote
computer.

rdavis@nyx.net
IRC Arvid