Subject: Re: SS20 network performance, take 2
To: None <port-sparc@netbsd.org>
From: Gregory Leblanc <gleblanc@linuxweasel.com>
List: port-sparc
Date: 05/25/2002 12:00:46
On Sat, 2002-05-25 at 10:33, Brian A. Seklecki wrote:
> On 21 May 2002 @ 4:55am (-0000), jwbirdsa@picarefy.com wrote:
> 
> >    Are you sure that the cards and hubs/switches on both ends agree on
> > speed and duplexing? These cards have a history of not autonegotiating
> > properly. It sounds like they're at least negotiating to 100mbps properly,
> 
> Right, an unbiased test would include a x/over ethernet cable of
> appropriate length between both machines (as to avoid NEXT) and both cards
> forced 100/full.
> 
> When testing, avoid using a switch or a hub (spoken: 'multi-
> port-repeater').  If there's a device sitting between the two machines,
> the test is skewed from the beginning.

How do you figure?  Modern bridges (Ethernet switches in marketing
speak) operate at full wire speed with impressively little latency
added.  The latency caused by using a modern bridge is a small fraction
of TCP overhead, and only a fraction of UDP overhead.  Also, modern hubs
are multi-port repeaters, but modern bridges (switches) do quite a bit
more than that.  

> If you have switch w/ management features, put both ports in thier own
> vlan as to consolidate the collision/broadcast domains.

This is only important if you have a semi-busy network.  On my local
lan, it'd make little or no difference.  

	Greg

> > or it wouldn't be working at all/achieving those speeds, but HDX/FDX
> > confusion can leave the link working, but limping along with poor
> > performance.

-- 
Portland, Oregon, USA.
Please don't copy me on replies to the list.