Mail Archives: djgpp-workers/2000/09/26/05:26:17
On Tue, 26 Sep 2000, Eli Zaretskii wrote:
> > > How much memory does that 386 screamer have? Is it possible that the
> > > large performance hit is due to paging because of the larger memory
> > > required by bnu2951b, rather than to actually slower code?
> >
> > 4 MiB.
That's a bit small. I take it you've taken the advice from the FAQ about
configuring DOS and DJGPP for optimal performance, for this amount memory?
> > I know that paging can make performance go out the window, but this is
> > a little extreme; if it only was three or four times as slow or even
> > ten, but ~65 times?!
>
> Paging can cause even 100-fold slow-down.
Or even a thousand. Just as a rough comparson:
data transfer rate to RAM: roughly 50 Megs/second
data transfer rate to HD : roughly 0.5 Meg/second (for 386's)
That already accounds for a factor of roughly a hundred. But we haven't
mentioned the worst factor, yet:
access latency of the HD : roughly 20 ms
worst case: random access to the page file, for each 4KB paged out:
--> 4kb every 20 ms (give or take another factor of two...)
--> 0.200 Megs/s
That's alreay a factor of 200 slower than main memory. Now imagine you
have a PIII-800 or so, with a RAM access speed of O(1GB/s), to compete
with, and that HD's haven't got more than about a factor of 2 better in
latency, since the 386's time...
> Does someone still have ld 2.8.1 installed to run a comparison?
I think I still have it. What's the exact test case?
Hans-Bernhard Broeker (broeker AT physik DOT rwth-aachen DOT de)
Even if all the snow were burnt, ashes would remain.
- Raw text -