early results 1.4b1 Solaris


Subject: early results 1.4b1 Solaris
From: Jon Piesing (jon@prl.research.philips.com)
Date: Thu Oct 03 1996 - 05:05:40 EDT


Well, I was so excited about the Netatalk 1.4 beta 1 supporting Solaris that
I went and installed it immediately. After some teething problems (machines
locking up) I managed to run the following performance comparison between
netatalk and CAP.

This test is copying a 75MByte file between a Mac (7500/100, 7.5.3, OT 1.1b12)
and a Sun UltraSparc 170E (Solaris 2.5).

First CAP :-

Sun to Mac : 3mins 20 secs

  PID USERNAME THR PRI NICE SIZE RES STATE TIME CPU COMMAND
  336 root 1 -3 0 1632K 496K run 673:27 35.35% uar
21262 jon 1 27 0 1784K 1352K sleep 0:16 11.76% aufs

Mac to Sun : 4mins 42 secs

  PID USERNAME THR PRI NICE SIZE RES STATE TIME CPU COMMAND
  336 root 1 26 0 1632K 496K run 674:26 25.35% uar
21262 jon 1 28 0 1784K 1240K run 0:47 16.08% aufs

Then netatalk 1.4b1 :-

Sun to Mac : 2 min 41 secs

  PID USERNAME THR PRI NICE SIZE RES STATE TIME CPU COMMAND
 2623 jon 1 17 0 1696K 1328K run 0:42 16.51% afpd

Mac to Sun : 4:10

  PID USERNAME THR PRI NICE SIZE RES STATE TIME CPU COMMAND
 2623 jon 1 27 0 1696K 1360K sleep 0:12 11.78% afpd

It looks like netatalk is coming in about 10%-20% faster and using a *LOT*
less CPU time (assuming TOP can be believed for time spent in the kernel doing
ddp). The Sun perfmeter utility also shows a lot less CPU used when working
with netatalk.

I'm going to try a faster Mac this evening.

It has to be said that even the best of these numbers (160 seconds) is
only 3.75 MBits/second which isn't that impressive for a lightly loaded
ethernet. That's probably just AppleTalk I guess.

Jon



This archive was generated by hypermail 2b28 : Sat Dec 18 1999 - 16:24:14 EST