Subject: Re: [netatalk-admins] 8.5 and asun
From: Abel Deuring (A.DEURING@BIONIC.zerberus.de)
Date: Mon Oct 26 1998 - 18:00:00 EST
dgraph@tiac.net wrote
> I'm not seeing any speed difference when using 8.5 to copy to the
> netatalk server.
>
> I tried it on a 8100/110 using an Asante 100 mb card.
>
> The computer has two hard drives so I loaded the os on the other.
>
> I boot it under 8.1 and I get about 20 seconds to send a 40 meg file.
> Under 8.5, it's the same.
>
> Is the current netatalk+asun taking advantage of the "improved" speed
> I'm hearing about or is this "improved" speed more fiction than fact?
Trying to carry out speed tests on the basis of simply copying a
file is somewhat questionable. This involves not only the speed
of the network part of the Mac OS but also the speed of the hard
disks (both client and server), the quality of the network
cables, the quality of the hub or switch, performance of the
server hardware and probably some other things I had the luck not
to get involved with.
I don't have an exact answer to your question but the following
might give you an idea how complex things can be.
Last weekend, I was able to make some performance tests with a
Sun Enterprise 450, 256 MB RAM, Solaris 2.6, three "striped" 18
GB Barracudas, mirrored to another three Barracudas, Helios
EtherShare 2.6, and a Linux Server (HP Server - I-dont-know-which-
type-exactly -, Pentium II 333(?) MHZ, 96 MB RAM, RAID-SCSI-
controller with five 7200 rpm IBM disks in RAID 0 mode), running
netatalk-1.4b2+asun-2.1.0-5 and SuSe Linux 5.3.
Running Helios Lantest on a recently delivered G3 Mac (I think it
has MacOS 8.1 installed but I'm not sure...) with a 100 MBit
Kingston card showed generally similar results both for the Sun/
Helios machine and the Linux/Netatalk box: read and write
transfer rates between 8 and 10 MB/s; the other parameters tested
by Helios Lantest were also nearly equal.
Helios Lantest is a - very useful - synthetic test, but it
measures only the performance of the "network parts" of the
client machine and the server (assuming the server is able to
cache 3 resp. 30 MB of hard disk data), so I added another test:
To copy a "folder hierachy" containing around 190 MB from/to the
servers. The results surprised me: copying from/to the Linux
machine needed between 2:10 and 2:30 minutes while the sun needed
2:50 ... 3:xx minutes. (times simply taken with a wrist-watch,
figures recalled from memory - I'm writing this mail from home
while the results are written down in the office. If there is
interest in it, I can post the exact figures)
So, while the "simple" Helios Lantest showed really good results
for both servers, the results of the "real" performance tests are
quite disillusionizing: 190 MB in 130 .. 180 seconds means a
transfer rate of 1 .. 1.5 MB/s - and the far cheaper Linux/
Netatalk machine did a better job than the Sun. (The conclusion
that it would be generally useless to use a Sun or HP or what
ever machine instead of a Linux server would nevertheless be
wrong - but that's a different discussion, involving the over all
performance of the hardware, what you need the hardware for,
availability of software packages, quality of support - for my
experience, newsgroups and mailing lists like this one are often
far better than commercial support -, availability of bug fixes
and many other things.)
I also run a low budget server (Linux/Netatalk/Samba P120 with
"cheap surroundings"), but it is used for special purposes - as a
reliable RIP (the combination of Ghostscript with Netatalk and
Samba is in the daily business far better than some commercial
bullshit), and as a seldom used image archive server. It is not
expected to be a high end server. Helios Lantest read/write gets
around 4-5 MB/sec; a simple copy command from/to its Quantum
Bigfoot image achive disks would probably give lousy results.
To come to an end: Try to isolate your problem. Look out for (or
write :) a program like Helios Lantest which generates data on a
client instead of copying the data from/to the client's hard
disk. Check the configuration of your client. Check your server
hardware: network card, speed of hard disks, speed/quality of the
SCSI (or IDE?) controller, of the network card, of your hub/
switch and of your cabling (can be really nasty work, the
latter...).
Abel
This archive was generated by hypermail 2b28 : Sat Dec 18 1999 - 16:33:39 EST