Messages in this thread | | | Date | Thu, 3 Jun 1999 17:12:11 +0200 (MET DST) | From | Bjorn Wesen <> | Subject | Re: zero-copy TCP fileserving |
| |
On Thu, 3 Jun 1999, Artur Skawina wrote: > > If your CPU needs to read 100 MByte and then write it back to ram, it does > > take twice the time as just reading it. > > no, Richard is right - on modern cpus the additional writes don't slow > you down that much (eg on pii - only ~7% for fullsized eth frames).
[trying not to flame!]
no CPU in the world has memory writes for free! i'll rephrase if its hard to understand - the _performance_ effect is not "twice the time" but relative to the cpu/bus speed, but the _absolute_ time spent is (+/- depending on bus design).
also, it doesn't matter if it's not noticeable on a modern P2, because we're not talking about P2's here and just because the 10/100 load on a P2 is ok now doesn't mean it will be tomorrow. grrrrr whatever happened to complexity theory in school...
> > Fact: our HW, while sending TCP on a 100 mbit/s ethernet, spends 70% of > > the time in csum_partial_copy. I'd be surprised if the performance of the > > transfers wouldn't go up quite alot if that call was eliminated. > > what hw is that?
an embedded network cpu, 100 mips. the cycles spent on copying are much better spent doing something useful.
[rant on]
the sort of thinking that goes "it's not noticeable on a $500 CPU right now so we don't have to care" is what keeps M$ programs slow and intel rich.
/bjorn
- 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/
| |