The state of 64-bit Desktop Linux

[Lennart Sorensen]

On Wed, Feb 11, 2009 at 04:34:34PM -0500, D. Hugh Redelmeier wrote:
> There is something odd about the slowdown you report.
> 
> Do you know why you feel that it is slowing down?
> 
> - Is the bottleneck CPU cycles, disk activity, network bandwidth, or
>   something else?  We have been inferring that you think CPU cycle are
>   the issue.
> 
> - if it is CPU cycles, where are they going?  The P4D is pretty fast.
>   In fact, CPUs are not getting much faster in recent years.  Just
>   more cores.  I don't find good machines from four years ago to be
>   slow.

The P4D also throtles if it overheats (which is does easily), up to 50%
down as far as I recall.

>   + is there heat-throttling going on?  The P4D is quite capable of
>     slowing down when it gets hot.

Exactly.

>   + is some useless process eating CPU resources?  top(1) might help you
>     discover this.  You can run a graphical system monitor program to
>     show you another dimension of this.  FWIW, I've had Firefox
>     quietly go into 100% CPU-eating mode.  X too.  Flash! might be
>     another villain (I don't have it).

Swapping makes things slow, flash spiing at 100%, firefox spinnning at
100%, running sched_idle tasks messing up the schedular (2.6.19 to
current seems affected by this very annoying and hard to fix bug).

>   + is some kernel activity eating CPU.  Most tools don't help you
>     figure that out.  I've seen that happen too.

Well top and iotop are both handy at times.  vmstat is handy too.

>   + are you on Jolt and expecting the CPU to get faster as you do? :-)
> 
> I recently had an interesting experience.  My hard drive was dying.
> It would cause system slowdowns (I think) because of retrying.  No
> symptoms but speed.  SMART scanning found some problems.  Touch wood,
> the system seems much better with a new drive.
> 
> That might be the case (pun intended).
> 
> 50W?  Do you mean 50W extra?  500W?

I suspect extra.

> I put a high-end nVidia card in my HP desktop a few years ago.  I
> could not find reliable information about what power was required.
> They would suggest an over-all system power supply rating.  Stupid
> since the suggestion took no account of other things in the system.
> They should specify the maximum power draw for the decice itself.
> 
> I was able to find out the HP power budget for add-in cards (good on
> HP).
> 
> Just to be safe, I installed a larger power supply (but no new fans).
> 
> I like HP desktops so far.  Lennart has higher/different standards.
> My HPs have been quite quiet.  But then I've never bought a P4
> (Athlons seemed always like a better choice).

I expect machines to not break.  I expect them to have BIOS updates when
needed (and not have buggy BIOS's in the first place).  HP has been bad
on both counts for me (OK, not emachines bad, but close).

> P4s do 64-bit really badly.  I think many 64-bit operations have to go
> through the ALU twice.  It was a tack on.  Athlon 64 and Core 2 are
> quite good at 64-bit operation.

Yeah it was a quick hack.

> 64-bit isn't as much of a win as one would expect.  Code density goes
> down so there is higher cache pressure.  I've run x86-64 on my desktop
> for three or four years but I don't think that I'd notice the
> performance difference (it takes a fairly large difference to be
> noticeable without some yardstick).

It certainly makes a difference to some tasks, but it does depend.  I
have seen 10 to 20% mentioned on many tests under linux, but it does
depend on the task.  There are the occational odd case where the cache
overhead actually hurts instead, although in the case of x86 with
variable length instructions, it is only actually pointer size that
hurts code size.  The instructions themselves are no larger than 32
code.

-- 
Len Sorensen
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists