Re: ext3 or xfs for desktop laptop
On Sat, 10 Jun 2006 18:05:20 -0400
"David R. Litwin" <presently42@gmail.com> wrote:
> On 10/06/06, Ron Johnson <ron.l.johnson@cox.net> wrote:
> >
> >
> > > Finally, the swap. I'm not too sure what these swap-files are,
> > > but it seems to me that a swap partiton is quite acceptable as a
> > > just-in-case. I'm simply unsure as to how large I should make it.
> >
> > A swap *file* does the same thing a swap *partition* does.
> >
> > Amazing, no?
>
>
> I understand that part. I do not understand why one would be better
> than an other. Nor
> have you indicated what guidelines I can use for determining an appropriate
> size.
>
Swap files allow more flexibility than swap partitions since you can change the
size if you do decide you allocated too much or too little. You can also add
swap files on the fly for the rare occasions where you need more swap.
They did have the downside that they used to be slower (someone here claimed
that this is no longer true for kernel 2.6). It was also not possible until
some time ago to use suspend to disk (or swsusp2) with swapfiles, only swap
partitions. That is also not the case anymore.
On the other hand, allocating swap area requires in-memory structures to keep
track of where things are, so more swap means less available physical memory
but more logical memory (if that is the right word). There is thus a trade-off
here. General it is advised to allocate about (1 - 3)*physical memory size,
although more memory usually means you need less swap.
> > As an aside, Mr. Johnson noted that hyperthreading can slow my
> > > computer down. In relation to the research that I did when I was
> > > wondering which kernel to use (a long time ago), I found out that
> > > hyper threading simply tells the processor to use it's free time
> > > to execute more activities. I failed to see (and still do) how
> > > this can slow my computer down.
> >
> > Read more.
> >
> > tells the processor to use it's free time to execute more
> > activities
> >
> > That's what *every* multi-tasking OS has done for the past 30 years.
> >
> > HT makes the CPU look, to the OS, look like 2 CPUs. Good in some
> > circumstances, bad in most.
>
On the one hand, HT allows better utilization of the cpu but allowing
instructions that need different parts of the cpu (different pipelines, like
int and float instructions) to run concurrently (finally ideas from sparc
systems are arriving to cisc, but they are much harder to implement under cisc).
The problem and advantage is that the cpu looks like 2 cpus (smp machine).
Different programs usually run faster this way since they are not synchronized
and a lot of times they actually do need different parts of the cpu. The
problem starts with scientific programs (or others that use mathematical
algorithms) which are designed for multiple processors and thus allocate several
threads that do pretty much the same thing. In such cases the problem tends to
start tripping over itself trying to utilise the same pipelines which it thinks
sit on two cpus but are actually are one pipeline on one cpu (I hope I wasn't
too unclear about this).
Whether HT is good for you depends mainly on the type of work you plan to do on
you computer (regular desktop use will generally benefit from hyper-threading).
>
> This is of particular interest to me. In what ways is HT "bad is most [ways]"?
>
>
Reply to: