Subject: performance comparisons with netatalk 1.4b1
From: Jon Piesing (jon@prl.research.philips.com)
Date: Tue Oct 08 1996 - 14:13:52 EDT
Dear all,
Various people got back to me after my simple performance comparisons between
cap and netatalk 1.4b1 (Solaris). In the light of these comments, I've done a
few more tests ....
As before the tests comprise copying a single 71M file between various systems.
In all cases the Sun is an Ultra 170E running Solaris 2.5.1 with Netatalk 1.4b1.
Here is a little table showing the results I got with a wide variety of machines.
A B A->B B->A Comments
7500/100 UltraSparc 4:10 2:41 what I got last time (OT 1.1b12 on Mac)
Quadra 650 UltraSparc 4:29 4:10 a 68K mac just to see
8100/100 UltraSparc 4:43 3:58 Power PC not running OpenTransport
8100/80AV UltraSparc 3:30 3:20 same as above, different network seg.
8100/80AV 8100/80AV 10:11 5:45 two hardware identical Macs,
(A) is same machine as used in previous line. The only reason we could find
for the *2 difference is that the disc on (B) is pretty full so we could
have a fragmentation problem there. These machines are on the same network
segment.
7500/100 8100/100 5:39 6:25 same 7500 as first test. same 8100 as 3rd.
7500/100 8500/120 9:00 4:27 same 7500 as first test. the hard disc on
the 8500 might be suspect.
Sparc 5/70 UltraSparc 1:13 1:17 two Suns for comparison, both Solaris 2.5.1.
done using 'cp' command over nfsv3. cp command executed on Sparc 5/70
Even if the 9 and 10 minute results are due to problems with the Macs concerned, I
would not have expected the Mac to Mac figures to be consistently slower than the Mac
to Sun figures. I'm also surprised that AppleTalk is so much less efficient than
nfs which itself is not the most efficent of protocols.
Jon
PS In case it makes a difference, our network is a number of coax segments connected
to a Kalpana etherswitch. The UltraSparc is connected direct to one port of that switch.
The Macs are connected to various ports. The test with the two 8100/80AVs was with two
machines on the same port.
This archive was generated by hypermail 2b28 : Sat Dec 18 1999 - 16:24:16 EST