Here are the results of some tests performed on the EtherTalk card and the
AppleTalk card. They suggest some speed advantages using the EtherTalk card
in a multiuser AppleShare application.
Some Hints:
The EtherTalk boards come preset for thick-wire EtherNet. To change them to
thin-wire it is necessary to move a jumper on the board. This procedure is
well documented in the EtherTalk Interface Card manual.
It is also mandatory that the cable itself be terminated on both ends.
Terminators must fit on the open end of the T-connection on each card. (The
testing group took one from their FastPath for these explorations.)
Also, the EtherTalk software must be installed into the system via Installer,
so that it shows up in the Control Panel as a Network icon on the left, which
-- when selected -- allows you to choose the built-in port or the EtherTalk
card. If you are connected to two networks, one AppleTalk and one EtherTalk,
you can toggle back and forth with the Network icon in the Control Panel.
This is not a gateway from one to the other, just a means of switching back
and forth.
The Benchmarks:
As part of the test, some benchmarks were performed which are documented
below. The numbers are representative of the differences, but will change
according to configurations, number of users, and the application being run.
- For one benchmark, Excel was opened on the server from the workstation in
both modes: EtherTalk and AppleTalk. The AppleTalk time was roughly 30 seconds;
the EtherTalk time roughly 19 seconds -- approximately 37% faster.
- Another test was running HyperCard from the server itself. EtherTalk seemed
to be at least 95% as fast at showing all cards in the address stack as
HyperCard would be from a local hard disk.
Finally, the most comprehensive benchmarks. Two sets of tests were run,
called Configuration A and Configuration B.
Configuration A:
- These were done on a Macintosh Plus-based system with a 4MB Macintosh Plus
as the AppleShare server, and a 2MB Macintosh Plus as the workstation. RAM
cache on the server was automatically set by AppleShare v1.1 at 32k.
(NOTE: Larger RAM cache settings actually slowed down the response times!)
The workstation was operating from a single floppy containing Finder 6.0b2 and
System 4.1. The application 4th Dimension (runtime version) was running off
the file server, since there wasn't room for the application on the floppy
containing System and Finder. 4th Dimension was running in multi-user mode
with one user. The network was AppleTalk, both cable and protocol. The server
used an Apple 40SC drive.
Configuration B:
- Tests were done on a Macintosh II based system with a 5mb Macintosh II as
the workstation and a 1mb Macintosh II as the server. Both Macintoshes were
connected using EtherTalk. The server disk was the same Apple 40SC as in
Configuration A. Again, the tests were done with the workstation using a
floppy-based System v4.1 and Finder v6.0b2. RAM cache settings were not
recorded. 4th Dimension was running in multi-user mode with one user off the
server volume.
The time tests involved two parts: timing the launch of the data base, then
timing a series of typical operations. These operations typically involved
reading and/or writing data, and the reading of screen layouts (mostly
dialogs).
The Results:
TEST CONF. A (seconds) CONF. B (seconds) % DIFFERENCE
---- ----------------- ----------------- ------------
Launch 359 124 35%
Oper.1 12 5 42%
Oper.2 26 10 38%
Oper.3 15 6 40%
Oper.4 28 11 39%
Oper.5 18 6 33%
Oper.6 35 12 34%
It appears from these tests that the Macintosh II in combination with
EtherTalk is two or three times faster than the Macintosh Plus with
LocalTalk cabling.