Messages in this thread | | | Date | Tue, 31 Mar 1998 07:35:30 -0500 | From | Raul Miller <> | Subject | Re: GGI Project Unhappy On Linux |
| |
Alan Cox <alan@lxorguk.ukuu.org.uk> wrote: > Moving some of the low level support doesn't change the stability per > se, but it does give you more control over things and especially over > abitration. The problem is there is a penalty for that. A syscall > costs time and you have to really get the syscall count down to sane > levels. A fast graphics card can plot a line in less time than a > syscall takes, so it is a real hard issue
Er.. I've not been following this whole thread (not enough time), but the kgi folks have had a solution for that one for quite a while:
declare a circular buffer somewhere in virtual memory that's large enough for at least two physical pages (probably want more). Keep one page unmapped and flush the mapped one(s) on segfault into buffer, or on explicit flush request.
There's still a bit of a latency penalty, and of course you need to use memory, but the syscall penalty is amortized over many requests.
-- Raul
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu
| |