The SPUs seem to become VERY irrealistic

Florian Hess (FLoH) hipabos at web.de
Sat Oct 19 03:58:39 PDT 2002


Hi,

Gerard Beekmans schrieb:
> On October 18, 2002 03:25 pm, Florian Hess (FLoH) wrote:
> 
>>measurement of 1 SPU = 892sec is no more realistic at all. The
> 
> 
> SBUs can't really be seen as linear numbers due to all kinds of external 
> factors (such as Matthias already explained). The SBU average still works out 
> in most cases but you have a very old system which is a lot slower than most 
> of the ones we tested SBU averages on. We're still talking about averages. 
> The point is, that for a faster system the averages will differ a few 
> minutes. But the slower a system gets, the bigger the average differences 
> get. For your machine 2 extra hours is about right
> 

Say which phisician is not dreaming of the very value to describe all 
things ;-) Maybe you should put a warning into your book, where you do 
postulate the SBU to be generally valid. I propose the last two 
sentences of the quote above.

> 
>>The dma mode has been activated at 10:00p, but he doesn't seem to 
>>accelerate the compilation.
> 
> 
> That means for 3.5 hours you didn't have DMA enabled. Not enabling DMA makes a 
> harddrive very slow. In my machine not using DMA hdparm shows a throughput of 
> 7 MB/sec and with DMA  46 MB/sec . That's quite a difference.
> 

On my other mashine, a p200/32m/20G, it's 6MB/s without and 9MB/s with 
DMA. The difference is not so big. Since the ram of my lfs system is 
even smaller, how would be the difference? You couldn't feel it even 
at the hd running sound (on the p200 that was possible).

> During a GCC installation a lot of files are written to your disk. If DMA is 
> disabled all those write operations take a lot longer.
> 

So I'll do another try *sigh*. By the way: where does the compiler 
store the compiled files before they are installed by 'make install'? 
On the host or on the lfs system? And where, there? Should I delete 
them before trying once more?


> Or try the next big package, Glibc, later on and see how it does (make sure 
> that you run hdparm to turn on DMA. I'd put it in a bootscript so you never 
> forget to do so).
> 

Eh? Glibc is installed in the chroot in ch6 - with the compiler 
already installed. No Glibc before compiler, or am I wrong?

The 32-bit block I/O (or so; enabled with c1) should also be enabled I 
guess?

Ah, here is some data:

<snip>
|pilote:~# hdparm -t /dev/hda
|
|/dev/hda:
| Timing buffered disk reads: 64MB in 23.52 seconds = 2.72 MB/s
|
|pilote:~# hdparm -c1 -d1 /dev/hda && hdparm -t /dev/hda
|
|/dev/hda:
|setting 32-bit I/O support flag to 1
|setting using_dma to 1 (on)
|I/O support = 1 (32-bit)
|using_dma   = 1 (on)
|/dev/hda:
| Timing buffered disk reads: 64MB in 19.40 seconds = 3.30 MB/s
|pilote:~# _
<snip>

What a difference! :-|

By the way this is a WDC33100H, capacity 3 GByte, acquired in a PC 
02/08/97

FLoH.

-- 
Unsubscribe: send email to listar at linuxfromscratch.org
and put 'unsubscribe lfs-support' in the subject header of the message



More information about the lfs-support mailing list