Mac Forum / General / Hardware / September 2004

It's mostly a testing methodology issue, somewhat exacerbated by the OS.
The CPU really isn't a problem here.

You should find other ways of measuring performance than transferring
files over what I'm guessing is the built-in AppleShare functionality.

It may be informative to divide these numbers down to the actual
bitrates.

For the 10M links:
493M/20min = 197.2Mbit/min = 3.3Mbit/s
165M/10min = 132Mbit/min = 2.2Mbit/s

For the 100M links:
493M/12min = 328.7Mbit/min = 5.5Mbit/s
165M/6min = 220Mbit/min = 3.6Mbit/s

Note that none of these numbers come close to the theoretical maximum
bandwidth of a 10M link (let alone 100M)

Of interest (but not surprising) is that the faster link speeds do
produce faster results, even though the 10M tests didn't come close to
actually moving 10M/s across the link.  This is actually a common
phenomenon - when upgrading a network from 10M to 100M to GigE, I've
seen file transfer speeds (between UNIX servers) move from 7M/s to
60M/s to 300M/s.

Common sense would tell you that you wouldn't gain a speedup from
boosting the link speed if you aren't hitting the limit at the lower
speeds, but it doesn't seem to actually work that way.

WRT why you see this, file transfer tests don't just test the network.
They also test your hard drive speed (on both sides) disk caching,
file system performance, operating system performance and many other
factors.  The same hardware running an operating system optimized for
efficient network file access (like some varieties of Unix and file-
server operating systems) may show dramatically different performance
numbers.

Neither of these machines are terribly fast, but both are plenty fast
for keeping a 10M link saturated.  I've run Linux-based router
software on much slower hardware than that that can keep up.

But MacOS 9's file sharing is not optimized for performance the way
an OS designed for file-serving would be, and the Finder is known to
not be a very fast file-transfer application.

It might be informative to run similar tests using dedicated
file-transfer programs (like an FTP client and FTP server) and see
how those numbers compare.

-- David

Much of it is accessing multiple small files (small being relative to
network speed, say up to 10MB). On an OSX machine running netstat to get
a dump of just how much data is moving per second, when copying multiple
files across a 100BaseT connection you can have transfer rates of a few
KB per second up to 8-9MB per second. It's really a case of a complex
interaction where the network itself isn't the bottleneck all the time,
sometimes it's the searching for small files on the sending end,
sometimes it's the network connection, and sometimes it's the other end
writing files.

If you had a single 2GB file and sent it over the network from one
unfragmented disk to another unfragmented disk, you would see a
sustained rate close to the 10MB per second, as that removes the drive
access from the equation. I get this when I .tar up folders to backup
via the network.