FreeBSD 4.x and OS-X tcp performance

[Bill Vermillion]

"Ang utong ko ay sasabog sa sarap!" exclaimed Charles Sprickman
while reading this message on Fri, Mar 04, 2005 at 18:43  
and then responded with:

> On Fri, 4 Mar 2005, Darcy Buskermolen wrote:

> >On Friday 04 March 2005 14:34, Charles Sprickman wrote:
> >>Howdy,

> >>Sorry to bring what seems like a simple issue up here.
> >>I had been blaming slow afp filesharing between my OS-X
> >>(10.3.8 and previous) and FreeBSD 4.x boxes on netatalk's
> >>afp implementation for some time. Not too long ago I got
> >>frustrated with this and tried smb and then ftp. On a simple
> >>10/100 network, and even with just a crossover between two
> >>boxes it seems that any tcp transfer tops out at around
> >>250KB/s.

> >>On the same network using the same switch I can get near
> >>Oline-rate to an penBSD box and to another OS-X box.

> >>If I use nfs and force udp as the transport, I *do* get near
> >>line-rate between OS-X and FBSD.

> >>My 5.3 box is tanked at the moment, so I cannot tell if the
> >>problem happens there as well. I do have a full ADC account,
> >>so I will be testing with the latest Tiger preview shortly,
> >>and the ADC access does give me a decent bug reporting
> >>facility if the fault lies within the OS-X tcp stack.

> >>I'm no tcpdump wizard, would anyone care to help me track this down?

> >I'd start with ensureing your nic's media options are properly
> >set (I've seen this exact behavior during duplex mismatches)

> Yep, I wouldn't have come here without checking all the basics.  I should 
> also add that given three machines in my standard config I get the 
> following results which will also help rule out cabling/speed/duplex 
> issues:

> os-x <-> obsd - good
> os-x <-> fbsd - bad
> obsd <-> fbsd - good
> os-x <-> os-x - good

I've seen this before.  A client had three OS/X machines in our
rack, one standard G4 and two Xrack devices.

One had severe problems connecting withour FreeBSD machines but not
to others.

I was never able to get with him in an interactive mode to check
things.  

He said he'd tried different settings but never got with me when he
was doing this so that I could check his machine, our machine, and
the intermediate switch.

He moved out to a rack of his own so I never did find out what was
wrong.

I do suspect duplex problems.   He was connecting to one of our
Cicso switches and Cisco has some extensive docs on some
configuration problems.  Part of it comes from when some vendors
decided to add their own features and violated standards.

http://www.cisco.com/warp/public/473/46.html

This is the link I saved from awhile back.  I hope it is
still valid.

But problems like this usually come from one side or the other not
properly responding to auto-negotiation as documented.  When
auto-neg fails at least one side will go to half-duplex, and with
the other in fdx then you usually only see throughput of
about 10% of normal because of data being sent back on a line that
the fdx line thinks is clear to send upon.

Bill


-- 
Bill Vermillion - bv @ wjv . com