Back to MacWindows home

(These tests were not conducted in collaboration with MacWindows. We are merely reporting on the presentation of the results at the Seybold Seminar.)

The authors said the tests were not designed to be all-encompassing. That is, the testers don't claim that these results will be the same in all setups and situations. For instance, they did not vary block sizes, and did not recreate the exact same network traffic for each test.

The Test Setup

There were two file transfer tested: a single 32 MB file and a 12 MB folder containing multiple small files. The Mac client was running Mac OS 8.1. The file transfer was initiated from the client Mac's Finder.

The network was 100BaseT (Fast Ethernet) with switched hubs.

The server software:	Windows NT 4.0 Services for Macintosh, Service Pack 3	SGI IRI 6.4 (UNIX), Xinet AFP server	Mac OS 8.1, AppleShare IP 5.0 and 6.0
The testing team	Walter Schild of Genex Interactive	Peter Brass of NewSub Services	Randy Wright of Anderson Lithograph
The server hardware	Dell PowerEdge	SGI Origin 2000	Apple Workgroup Server 9650
Server processor	400 MHz Pentium II (100 MHz Bus)	Two R-10,000	350 MHZ 604e
Server RAM	256 MB	512 GB	128 MB
Server Storage	40 GB RAID 3	400 GB RAID 5	18 GB RAID STRIPE
Number of users logged on	20	32	27

The Results

Schild's testing team found Windows NT 4 SFM to be significantly slower than both AppleShare IP and the Xinet on UNIX solution. NT took 60 percent longer than AppleShare IP 6 to write a 30 MB file over the network. NT took almost twice as long than AppleShare IP 6 to write a 12 MB folder.

AppleShare IP was just slightly slower than the UNIX server, even though the SGI server had two processors, each of which was faster than the 604e in the Mac. AppleShare was slightly faster in one test, writing a 12 MB folder. The team also found AppleShare IP 6.0 to be about 10-to-15 percent faster than AppleShare IP 5.0.

The team also found that AFP (AppleShare) over IP was up to 3 times faster than AFP over AppleTalk on the SGI server.

Finally the team compared 10BaseT with 100BaseT. Fast Ethernet moved files from twice as fast as 10 Mbit Ethernet (to read the 12 MB folder) to over 6 times faster (to write the 30 MB file).

A graph of the results should be posted at the Seybold Seminars' web site some time after Monday, September 7.