My two cents on network performance

Christopher A. Smith (casmith@clark.net)
Sat, 29 Jun 1996 13:52:27 -0400 (EDT)


At my workplace we have a mixture of AIX, HP, Solaris, and Linux systems
-- an increasing number of machines freed of their Windows 95
imprisonment, and a cheap and comparable alternative to their more
expensive HP, Solaris, and AIX counterparts. These Linux systems hold
their own quite nicely in many areas and, in some cases, outperform the
more expensive systems.

Linux started appearing in our development area last October. The
systems, Micron Millenium's with Pentium 133s and anywhere from 16 MB to
64 MB of RAM and IDE hard drives, were running Linux 1.2.13 and performed
beautifully. Our operations staff became very interested in Linux because
of the relatively cheap cost of Linux systems, compared to HP, Sun, and
IBM systems, and our developers became interested because they could
easily do their work at home without having to fight with modem servers.
I originally had an HP 9000/712 on my desk and a PowerPC 601 AIX system at
home, both of which I gave up for Linux because it worked better. We even
had a FreeBSD user switch to Linux. The only problem we encountered was
with NFS writing which was very sluggish from the perspective of the
user.

We're now running Linux 2.0, and the same sluggishness exists. To get a
better idea of how Linux was performing against the HP, Solaris, and AIX
systems, I did the following unofficial test. Each system was given a
local file -- not copied from the NFS server first, but taken from
something like /usr/bin, /sbin, /usr/sbin, /usr/lib, etc. -- which was
then copied to an NFS-mounted directory. This was done one early morning
when the few people who were at work were just reading news or mail or
talking -- not doing any work. The 16-megabit token ring network was very
quiet, and the NFS server, an IBM RS/6000 590 running AIX 4.1.4, wasn't
doing anything. Each client was basically asleep, too, doing nothing to
provoke the hardware. Here's what I found:

System Size of file Time Bytes/second
HPUX 9.05 on HP 9000/712 548053 1.07 512199
Solaris 2.4 on Sparc 5 543184 1.03 527363
AIX 4.1.3 on PowerPC 604 544525 1.24 439133
Linux 2.0 on Pentium 133 539398 12.77 42239

(The time was found by doing "time cp <local file> <NFS directory>")

The Linux machine, which has 64 MB of RAM, had just been rebooted and
wasn't running extraordinary -- just getty's, inetd, kerneld, syslogd,
etc. and ypbind. No X windows, nothing like that. just to be sure that
the slow write with Linux hadn't occurred during some network hiccup, I
did it again several times and got the same results -- between 40,000 and
43,000 bytes per second. (The HP machine has 64 MB of RAM, the Solaris
machine has 48 MB of RAM, and the AIX machine has 80 MB of RAM, by the
way. The HP has SCSI disks, the Sun has SCSI disks, and the AIX system
has IDE disks.)

Normally, our Linux users are dealing with small files or are working on
the local disk, so NFS is hardly a variable. However, when we try to
compile our project software on a Linux system, whereby the object files,
libraries, and executables are written back to the NFS server, the delay
is quite noticeable. Or when a user quits out of Netscape, and it fiddles
with the cache directory, which is done via NFS, the delay is VERY
noticeable. We began the practice a long time ago of compiling the project
software locally and then moving the objects, libraries, and executables
back to the NFS server with ftp or something similar.

I've thought about potential hardware problems and just don't see what it
could be. When the project software is compiled locally, the compile is
VERY quick, competing with or even beating the HP, Solaris, and AIX
systems. NFS reading and other network activities seem -- from the user
perspective -- to be just as fast as our other systems. It's just NFS
writing which is the bottleneck, from the "data" above and the user
perspective.

I can't imagine that Linux is that slow on NFS writing; I can't imagine
that the developers would tolerate it. If anyone has any ideas or feedback
on this, I'd appreciate it.

--------------------------------------
Christopher A. Smith
Arlington VA * casmith@clark.net