Subject: Re: netatalk performance
From: Rich Sudlow (rich@samson.cc.nd.edu)
Date: Thu Feb 01 1996 - 07:47:33 EST
>
>On Wed, 31 Jan 1996 16:31:44 -0600 (CST), Bradley Carlson
><carlson@sci.mus.mn.us> wrote:
>
>
>>Tests today done on almost-idle network, PM6100/60 as client.
>>
>>Copying a large Netscape Cache folder (average file size 7.1 Kbytes):
>> to Netatalk server: 4.6 Kbytes/second
>> to Mac : 5.8 Kbytes/second
>>
>>Copying a single large tar file (18,680 Kbytes):
>> to Netatalk server: 242.6 Kbytes/second
>> to Mac : 233.5 Kbytes/second
We've recently done some extensive testing (> 200 hrs) using the netatalk
xlator and AFS and have found that you can get very reproducable data
with large file's (5 MB) showing speeds > 500 KB/sec for write and 600 KB/sec
reads. We also found that that speed of netatalk is GREATLY dependent on
the speed of the client. ( A PM 6100 is a poor client..Try using a 68k
based machine e.g. PB 520 or Quadra 800 if you want to see some better
numbers..approximately 40 percent...Isn't native code great ;-) )
Also we found that you really want to shove as much RAM into the
xlator as possible for the best performance (this greatly speeds up
reads from the xlator..we're currently using 512 MB in a SS20).
(We didn't use the memory for an AFS memcache however)
We found that in general the network connection (and yes appletalk)
is the limiting factor of performance. Fast ethernet helps some to
for supporting a larger number of clients.
>>
>
>It appears that netatalk's weakness (or is it AppleTalk's weakness) is
>in opening and closing files then.
>
>It would be interesting to see some figues comparing copying
>Mac->Netatalk and Mac->Mac...
If you really want to compare numbers make sure that you do this on an
isolated subnet. You'd be surprised what a little network traffic will
do to consistency of your testing.
Rich
Rich Sudlow
Office of Information Technology
University of Notre Dame
This archive was generated by hypermail 2b28 : Sat Dec 18 1999 - 16:23:57 EST