Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 9 Mar 2005 03:27:06 +0100
From:      Bernd Walter <ticso@cicely12.cicely.de>
To:        Kirk Strauser <kirk@strauser.com>
Cc:        freebsd-alpha@freebsd.org
Subject:   Re: Question about cc flags in buildkernel
Message-ID:  <20050309022705.GF22167@cicely12.cicely.de>
In-Reply-To: <200503080950.26844.kirk@strauser.com>
References:  <200503072107.13313.kirk@strauser.com> <20050308034852.GX22167@cicely12.cicely.de> <200503080950.26844.kirk@strauser.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Mar 08, 2005 at 09:50:23AM -0600, Kirk Strauser wrote:
> On Monday 07 March 2005 21:48, Bernd Walter wrote:
> 
> > There is no speed influence by this option. 
> 
> Thanks for the explanation.  I wasn't sure, and it was nice to get a bit of 
> confirmation.
> 
> > How did you compare speed?
> 
> It's more of a general observation: everything is just slow, slow, slow.  
> For example, loading the man page for 'zshall' (from the shells/zsh port) 
> takes most of a minute:
> 
>    # time man zshall > /dev/null
>    Formatting page, please wait...Done.
>    man zshall > /dev/null  47.39s user 0.99s system 93% cpu 51.523 total
> 
> during which troff and grotty are using 100% of the CPU.  SSH connections to 
> it take a long time to start:
> 
>    $ time ssh gopher exit
>    ssh gopher exit  0.04s user 0.03s system 0% cpu 7.675 total

I'm not extremly surprised by that numbers.
You have a really old machine and gcc doesn't do very well with byte
oriented source on non BWX alphas.
But it looks even slower than on my NoName, which is definitively less
powerfull.
Under which FreeBSD version is this?

> If I run that with "-v", I can see that it spends most of that time in 
> "Entering interactive session."  The hard drive isn't completely horrible 
> (~8MB sustained transfers at less than 8% CPU usage), and it has 3 fxp NICs 
> that barely register under heavy network load.  I installed 256MB of RAM 
> (with 2MB of L2 cache, I think), and it's currently only 12KB into swap.  
> In other words, by every metric I can think of, it appears the bottleneck 
> is that the CPU is dog slow.
> 
> > If it's just from compile time, you shouldn't forget that gcc-3 is much
> > slower than the older gcc and that compiling for alpha is a much harder
> > job than compiling for i386.
> 
> Compile times don't really bother me; I launch big jobs and then walk away 
> from it until they're done.  That's also why I was using the higher-order 
> optimization flags.  They make the compiles quite a bit slower, but the 
> results usually seem to be worth it (which here means "less painful").

I don't think O3 is really that usefull - I typically just use O2.

> I'm really not expecting miracles from this little machine, but my K6/333 
> laptop with less RAM and a 2.5" IDE harddrive smokes it in every way.  I 
> was sort of hoping I'd found the "aha!" detail that would account for the 
> awful performance.

Well - a K6/333 is a few years younger than your alpha.

-- 
B.Walter                   BWCT                http://www.bwct.de
bernd@bwct.de                                  info@bwct.de



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20050309022705.GF22167>