On Wed, Dec 18, 2002 at 12:25:09AM -0600, Peter Seebach wrote:
> In message <20021218055605.5912BB@proven.weird.com>, Greg A. Woods writes:
> >I'm trying to say that those people punishing their machines with a
> >combination of CPU and I/O intensive work while at the same time running
> >a GUI and big GUI applications simply SHOULD NOT expect good interactive
> >performance.  They are getting exactly what I would expect the should
> >get.
> 
> I don't see why.  My Amiga could solve this problem.  My NetBSD 1.5 box
> solved this problem better than my -current system does.
> 
> >I'm also suggesting that if they demand good interactive support
> >ultimately then they must find some way to separate their job mix onto
> >at least two appropriately tuned systems.
> 
> I disagree, simply because I've seen better performance on this.  I mean,
> even *Windows* gives me better interactive performance under load right now.

And you don't need X to have problems (as it seems that X is being blamed
for all of this).  I have a CVS/NFS server (NetBSD 1.6, 700 MHz Athlon,
512 MB RAM, software RAID with 120 GB WDC disks (8 MB caches) and a 100 Mbps
Ethernet card) which has no interactive users, it's just a server.  If
someone's doing a cvs update from the server, I start getting these on the
clients:

Dec 18 08:58:50 pyy /netbsd: nfs server kyyhky:/home: not responding
Dec 18 08:58:55 pyy /netbsd: nfs server kyyhky:/home: is alive again
Dec 18 08:59:35 pyy /netbsd: nfs server kyyhky:/home: not responding
Dec 18 08:59:41 pyy /netbsd: nfs server kyyhky:/home: is alive again

The poor machine can't even handle one CVS operation without failing to
serve files over NFS.  CPU usage on the server is less than 20% and
there's over 300 MB of RAM available.

> I am strongly suspicious that there is a specific thing which will later
> be seen as "a bug" which is dwarfing the tuning issues right now.

Well, it can't be right that the machines of today are doing worse than
those 15 years ago..

Next they'll blame it on IDE, sure, but I don't buy that, either - I've
got only one disk per IDE port and the disks are using Ultra/100 DMA.
The disks are fast (but RAID appears to kill performance).

I'm also seeing the problems in interactive use on client systems, I'm
just saying you don't need X to notice these problems.

  -jm