Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 27 Jan 2000 20:57:06 -0700
From:      Warner Losh <imp@village.org>
To:        Juergen Lock <nox@jelal.kn-bremen.de>
Cc:        freebsd-stable@FreeBSD.ORG
Subject:   Re: Warner's PCI Modem Driver 
Message-ID:  <200001280357.UAA51783@harmony.village.org>
In-Reply-To: Your message of "Fri, 28 Jan 2000 04:26:15 %2B0100." <200001280326.EAA15051@saturn.kn-bremen.de> 
References:  <200001280326.EAA15051@saturn.kn-bremen.de>  <Pine.LNX.3.96.1000127082044.30216B-100000@mail.clones.com> 

next in thread | previous in thread | raw e-mail | index | archive | help
In message <200001280326.EAA15051@saturn.kn-bremen.de> Juergen Lock writes:
: They perform better??  shouldn't the isa bus' bandwidth be more
: than enough for the (even compressed) 56kbps of a modem?

We've found that we get better ping times (about 10% faster)[*], fewer
dropped characters due to interrupt latency[**] and the system seems a
little snappier[***].  We've also measured a 5% increase in ftp performance
on the usual ftp a big honkin' file.[****]

These tests were performed on an AMD 5x86 133 based machine (a 486
class machine running at a 33MHz bus).  The old ISA modems were US
Robotics Sportsters while the new PCI modem was the ActionTec FAX
Modem (don't know what they called this in their released version,
since this was a preliminary one).  We have about 50-100 firewall
rules (same ruleset for both runs, we've changed it since the tests I
performed, so I don't have an exact number).  The remove modem was a
US Robotics sportster internal ISA modem.

On the ISA based modems, we'd see about 10 dropped character events a
day when traffic got heavy.  We have two 33.6 connections on this
machine.  With the PCI modems, we see about 1 a week.  My guess is
that we're running right on the harry edge of needing more CPU power
and the lower PCI latency and faster bus response makes the
difference here.

So it isn't a huge win (you won't get 2x the performance), but it did
allow us to make our under powered machine last a few more months
while we saved for larger iron to replace it and two other machines
with (mostly to consolidate rack space given today's insanely
inexpensive machines).

While one external modem vs 1 internal modem may be a no brainer, it
gets really hairy when you have 5 or 6 external modems, each with
their own wall wort.  Those wall worts eat powerstrip real estate like
you wouldn't believe.

Warner

[*] 170ms vs 140ms, which is more like 20% is what I'm measuring now
on a mixed machine (we just brought up a new member and haven't
converted him to a new pci modem yet) right now.  I seem to recall it
went from 160ms to 140ms

[**] Haven't seen a buffer overflow since Dec 2 with the sio.  Before
we'd see 10 messages a day of xxx more interrupt-level buffer
overflows.  These were worst when large transfers were happening
during daily or weekly running.

[***] Purely subjective

[****] Typical ftp the kernel to /dev/null over the link 3 times,
average rates and compare before/after.  I can no longer find the
numbers to back up this claim, and I don't have an account on the
other end of new client to test it out via the same machine.


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-stable" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200001280357.UAA51783>