Re: mmap() versus read()

David S. Miller (davem@dm.cobaltmicro.com)
Wed, 11 Mar 1998 00:16:00 -0800


Date: Wed, 11 Mar 1998 01:31:43 +0100 (CET)
From: Thomas Heide Clausen <voop@innocent.com>

For your information: Solaris pukes in processes with more than
about 20000 concurrently active threads (I can repeatidly on a
Enterprise 2 with 1GB ram get it to puke on thread no. 14287).
Haven't figured out why yet, though....but it would be cool
if the Linux version didn't have such constraints....

If you got rid of the NR_TASKS and the task[] array limitations
(simple to do for the kernel on non-Intel platforms), Linux
on the UltraSparc could feasibly do this.

We create tasks close to 10,000 per second (real processes, not
threads), and my current TCP can setup and tear-down 2,500 sockets per
second (scaling down to about 2,100 per second once you build up
40,000 or so sockets in the system). Sounds like potential for a kick
ass web server to me ;-)

Later,
David S. Miller
davem@dm.cobaltmicro.com

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu