Messages in this thread | | | Date | Fri, 24 Dec 1999 12:05:01 +1100 | From | Lincoln Dale <> | Subject | Re: Bloat? (khttpd) |
| |
At 19:33 23/12/99 -0500, Mark Hahn wrote: >so far, we have no reason to believe that khttpd performs better than, >say, phhttbd, even on silly static-only benchmarks. and even if it did, >the sensible conclusion would be that there's something wrong with Linux, >not that webserving should be in the kernel!
actually, khttpd does get around one limitation currently inherent inside linux -- and that is that there is no mechanism for zero-copy.
a user-space application needs to read() from kernel-to-user on a socket, then write() user-to-kernel back.
2 x memory-copies.
i'm currently working on this; i have a _real_ application that cannot push more than 320 mbit/sec to/from the network (on gig-E).
its called too-many-copies and a 133mhz FSB.
we're exceeding performance levels that i suspect either of khttpd or phttpd have been benchmarked to. (ie. phttpd is mostly showing the benefits of async-i/o-event-notification. even it will be limited by copies in the long-run).
cheers,
lincoln. PS. rather than bitching and moaning about it, yes, i am working on a fix. the way i'm doing it is a new character device driver which allows user-space to mmap() into physical ram. maybe not pretty, maybe not acceptable to the l-k folk, but it will certainly get us out of a current ceiling limitation.
- 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/
| |