Date: Mon, 5 Apr 1999 18:04:21 -0500 (CDT) From: Jonathan Lemon <jlemon@americantv.com> To: jmb@hub.freebsd.org, freebsd-advocacy@freebsd.org Subject: Re: Very Interesting user of FreeBSD Message-ID: <199904052304.SAA11827@free.pcs> In-Reply-To: <local.mail.freebsd-advocacy/19990405222038.F2DDD15583@hub.freebsd.org> References: <local.mail.freebsd-advocacy/199904052038.PAA11647@free.pcs>
next in thread | previous in thread | raw e-mail | index | archive | help
In article <local.mail.freebsd-advocacy/19990405222038.F2DDD15583@hub.freebsd.org> you write:
>>
>> I was talking to the Polyteam at the conference, and what they would
>> primarily like is for my changes to be folded into FBSD (which I should
>> submit for review sometime soon). Also, they are still using 2.2.8
>> instead of 3.1, with one of the primary reasons being that the two-floppy
>> boot disk is a pain in the butt.
>
> appendix K of the report contains the configuration changes
>and one code change that they (the polyteam) made to their FreeBSD
>machines. interesting reading.
>
>kernel config:
>options "MAXDSIZ=(512*1024*1024)"
>options "DFLDSIZ=(128*1024*1024)"
Yes, these aren't needed on 3.1, as they've already been rolled
into that distribution.
>options "NMBCLUSTERS=40960" !!
Heh. I had an equivalent sized area, and still ran out of space
sometimes, (primarily because we fell behind in processing).
>options TCP_ACK_HACK
In 3.1, it is the sysctl knob "net.inet.tcp.delayed_ack".
>/sys/netinet/tcp_timer.h
>#define TCPTC_MSL (3*PR_SLOWHZ)
This violation of spec was needed to prevent the test boxes from freezing
up under the workload; at an average of 300 connections/second per test
pair, the the number of sockets in TIME_WAIT grows intolerably large.
I have an alternate implementation that moves all sockets in TIME_WAIT
state to a different list, so they don't impact normal processing.
>/sys/kern/uipc_socket.c: (faked diff)
> if (resid >= MINCLSIZE) {
> MCLGET(m, M_WAIT);
> if ((m->m_flags & M_EXT) == 0)
> goto nopages;
> mlen = MCLBYTES;
> len = min(min(mlen, resid), space);
> } else {
>+ atomic = 1;
>nopages:
This one I'm not sure what to do with. That is, I know what it does,
but am not sure what the correct fix is.
Anyway, Polyteam doesn't mind having to re-apply the diffs above
to the kernel, since they need to re-build the kernel with a lower
MSL and tweaked values anyway. They have stated (as I understand it)
that one of their current goals is to make available a pre-patched
FreeBSD kernel from their website so that their various customers can
just download and install a tuned kernel.
--
Jonathan
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-advocacy" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199904052304.SAA11827>