Re: hdparm to increase performance
On 10 Nov 2000, Krzys Majewski wrote:
> "Michael P. Soulier" <msoulier@storm.ca> writes:
>
> > In case you guys missed this one, check it out.
> >
> > http://www.oreillynet.com/pub/a/272
> >
> > I just increased my harddrive throughput by 5 times.
> >
> > Mike
>
> Do you notice a difference though? I increased mine from about 3Mbps
> to about 18Mbps, but I haven't felt it yet. -chris
Linux does not swap enough. Remove some RAM :-)
More seriously, I guess that if Linux has lots of RAM to play with
(here swap in use <10% RAM), and a big cache (30-50% RAM), you're not
going to saturate your disk bandwidth in the first place. So increasing
it by a factor of 6 is not going to help since it was not the bottleneck
in the first place.
Also under such conditions whether reads or writes are slow is not
really important since reads are read ahead, as long as the data arrives
before you need it it's fine, and writes go to the cache and don't have
to be flushed all that often.
Finally the only cases where disk access will have an influence is
when the read-ahead occasionally fails. In that case It's most likely
going to be a latency issue, i.e. what's your seek time. And I believe
hdparm cannot help enhance this aspect.
In such a situation, a more interesting aspect of hdparm is to reduce
the CPU utilization. That way gcc and others can continue to work while
you're accessing the disk. I've seen the usage drop from 75-90% while
doing a 'dd if=/dev/null of=foo', which made the machine very unusable,
to less than 5%, which made the machine usable again.
Of course it depends on your particular situation and usage patterns.
Maybe you stream tons of data, copy huge files (50MB+), etc.
--
Francois Gouget fgouget@free.fr http://fgouget.free.fr/
Linux: the choice of a GNU generation
Reply to: