[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-kernel
Subject:    Re: [patch] fastcall-2.3.32-B6, SYSENTER/SYSEXIT support
From:       Artur Skawina <skawina () geocities ! com>
Date:       1999-12-15 11:39:08
[Download RAW message or body]

Pavel Machek wrote:
> 
> Well, read from cached file is _not_ meaningless, still Ingo seen
> improvements on that.

it only matters if you use very small chunks (for 1 byte reads/writes
a 20+% difference is not uncommon), but already at 4096 bytes the
improvement is in the noise (while it may be measurable, i doubt you'd
notice it in rl). Where the lower read/write cost is more likely to
matter is ipc/sockets/pipes etc (when you either pass things around
or write small elements, like headers).


> I think you should concentrate on read(), write() and select()
> syscalls. Those are the ones done often.

it's not about often used syscalls; it's about speeding up _all_
of them. The only reason gettimeofday is singled out is because
while you get a nice 31% improvement, you also get new infrastructure
that lets you save another ~30%. [1] I can't think of other syscalls
this is true for right now...

[1] 1st number is measured difference, 2nd is estimation.


-
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/

[prev in list] [next in list] [prev in thread] [next in thread]