>>> No, that's not a fair comparison. Paging/swapping is the only way to
>>> extend virtual memory space without buying more RAM. Is there an
>>> alternative method of increasing VM size without paging, and is it any
>>> faster?
>
>> Sure it is fair. You could use overlays instead, but most people
>> are too lazy to do it. Overlays can give you better performance.
>
> There just isn't a 1e4 factor difference between paging and overlays,
> I'd be surprised if you get more than a factor of 2.
You can do very well. Start your IO early and asynchronously, so that it
completes just as you need the code/data. The OS can not predict this.
> And you could also tune
> your program to be paging-friendly by clustering functions/data used
> together... but I'd guess much of that happens semi-automatically by the
> way source programs are laid out (you don't usually define totally
> unrelated functions in the same source file, for one).
You can tune your program to be DSM-friendly by clustering your
access to the DSM. Much of that happens semi-automatically by the
way people write code. (you don't usually access totally unrelated
data -- if you do, nothing will save you)
>>> DSM can destroy performance for typical threaded programmes.
>
>> It can. It does not always. One might as well try, especially when
>> there is existing code that can be used.
>
> That it can _for typical programs_ means you have to be avare of the
> pitfalls... and that kills the "nice, friendly, uniform model"
> idea that is quite critical to the whole bussiness at hand.
So people should just be happy writing new code, since it _might_ be
very slow? That could be an expensive waste of time.
I'm sick of seeing this:
"Our way is the only way."
"You must throw out your inferior code. Do not attempt to reuse it."
"There must not be any possible performance loss, even if you don't care."
"We write manly code. You write girly code."
"Everyone must write kick-ass code, whether they need it or not."
"DSM is a sin. Sinners must repent!"
"Those that don't use every last cycle are idiots that must be saved."
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/