Summary - Hubs versus Switches

From: Koonz, Jay <jKoonz_at_USCO.com>
Date: Thu Jun 28 2001 - 16:59:55 EDT
Here's the original question - The replys follow. Thanks to one and all. 

============================================================================
=================
We have a dozen Solaris-7 or 8 servers hooked up via a bunch of hubs. We're
going to be moving them into a new computer room where they will all be
connected to switches. In the dim recesses of my mind, I seem to remember
some issues connecting Solaris boxes to switches instead of hubs. Can
someone help me out here, will my boxes know the difference ??
============================================================================
=================
Boxes won't know the diff, but autonegotiation can be an issue sometimes
(althouhg I haven't had any personally).  The big drawback is sniffers,
they don't work well on switches, other than that they rock.
						Jeff Kennedy
[jlkennedy@amcc.com]
============================================================================
=================
You'll want to make sure you set the machines to 100/full duplex in
/etc/system, as well as set the switch to not autodetect, but rather
100/fd.

Some switches don't cooperate with Sun boxes too well... most folks
don't even try; they make the settings after they install the OS.

Drop this into /etc/system on your hosts (assuming you have hme's)

* Begin hme - force 100BaseTX - Full Duplex Mode
set hme:hme_adv_autoneg_cap=0
set hme:hme_adv_100hdx_cap=0
set hme:hme_adv_100fdx_cap=1
* End hme

Or run from the command line to do it manually (one-time only,
is not persistant between boots):

ndd -set /dev/hme adv_autoneg_cap 0
ndd -set /dev/hme adv_100hdx_cap 0
ndd -set /dev/hme adv_100fdx_cap 1
						Tim Yocum [tim@playboy.com]
============================================================================
=================
If they are cisco switches, and have autonegotiation turned on, then your
solaris boxes may have a problem.  Set the speed and the duplex on BOTH the
switch and the solaris box to get the throughput you expect.
						mcferren@colltech.com
============================================================================
=================
	The only issue we had was many years ago; several of the
earlier "inexpensive" breeds of switches (NetGear, for one, comes to
mind; there were others too) had trouble auto-negotiating with the hme
devices.  On several boxes we had to force speed and duplex in the
/etc/system file.  We've not seen this recently, though; I'm not sure
if the newer switches got it right, or if later versions of Solaris
(we had this trouble with 2.6 mostly) had a workaround.  We never saw
this with larger switches, just the smaller "workgroup switches".
						John Sullivan
[jgs@perifitech.com]
============================================================================
=================
Should not be a problem, Sun NIC's are usually well behaved and will sync
correctly. Watch console messages and study the ndd man page for full
duplex and 10/100 autosensing information.
						Jed Dobson [jed@wgtech.com]
============================================================================
=================
My experience is that auto sensing switches never negotiated properly.  They
always negotiated to a less than optimal speed so we set either the kernel
using ndd or placed the appropriate commands in /etc/system to force the
server to the proper speed.

If you don't experience this problem then you probably shouldn't have any
problem.
						Richard J. Buchman
[rbuchman@netcom.com]
============================================================================
=================
All networking configs being the same, the switch just provides better
performance. Hubs arbitrate all boxes through a single hardware connection,
switches give added hardware functionality that give a non arbitrated
connection. Each box is a hardware connection end to end. Not having to
arbitrate for network banwidth is the performance enhancement.

There are issues with sun  and autonegotiate not working real well with
some Cisco equipment. But related to this issue, whether a hub or switch,
it makes no difference, there have been issues with both. You can read some
of the traffic on the list by searching the old database for ndd, or
autoneg.
						Steve Hastings
[stevehas@us.ibm.com]
============================================================================
=================
you might want to make sure they come up as full duplex if they are
capable.  use ndd to check this ...

# ndd -set /dev/hme instance 0
# ndd -get /dev/hme link_mode
1

a one back is full duplex (i chose hme0 here by using instance 0)
						Chris Cariffe
[chris@zippy.baydat.org]
============================================================================
=================
Some switches have problem autonegotiating with some Sun cards. I suggest
you avoid the issue by hard-coding all your Suns and switches to 100/full,
assuming that's what you want to use.

Attached is a script to make this job easier.
						Andrew J Caines
[AJCaines@borg.uunt.net]
============================================================================
=================
Mine didn't.  The problem you may remember was with gigabit cards and the
switches.
If you've worked through setting up a gigabit card then you know everything
you need -  It's still painful to set these gigabit NICs up.

Oh.  You'll love being switched.  Much improvement.
						Mike Peppard
[mpeppard@impole.com]
============================================================================
=================
As far as the Sun is concerned it should not care whether you are connecting
to a switch or hub. The key difference is what it does with the packets
after they have left the Sun. The main problem will be if the switch is
correctly setup, which is not a Sun problem.
						Peter Stokes
[peter@ashlyn.co.uk]
============================================================================
=================
Assuming we're talking 100BaseT switches and hme/qfe interfaces on the
servers, then the usual problem is failure of the host/switch pair
to auto-negotiate duplex.

Since this *usually* works, you can just bring everybody up, force some
network load for a few minutes, and check netstat -i.  Duplex mismatch
shows up as errors and/or collisions (all of which should be *zero*
in a normal switched setup).

If you decide (or need) to force full-duplex, the sunmgrs-faq details
how to do that (plus a summary in the last few days).  That's on the
host side, of course.  If you force the host, you should also force the
switch port, and that is vendor/model-dependent, of course.

						Jay Lessert
[jayl@accelerant.net]
============================================================================
=================
The biggest difference I can think of is that the network will be
much faster on the switch.

Otherwise, the Sun's won't care.

Make sure they are either all configured to auto-negotiate or that
you configure them (switch port and Sun network interface) to have
the same speed and duplex.

Suns seem to have issues with negotiating to 100 Mbit Full duplex,
but if you hard code both the Sun and the switch port to that, they
tend to run really well.

We have always used Cisco switches and never had a problem with the
auto-negotiation (except that they never went 100 full).

Hope that helps.

p.s. It may take a few seconds for the switch to map to the servers
(Sun or otherwise).  This is because the switches should be building
arp-to-port tables that let it know where the systems are plugged
into.  Once they are plugged in, you should not swap cables as this
will cause a small hesitation (5 to 10 seconds) while the switch
"relearns reality".
						Kevin Colagio
[kdc@rochester.rr.com]
============================================================================
=================
No differences I know of.  In general, 10/100 capable Suns have
trouble auto negotiating speed/duplex with any networking device
that is also 10/100 capable.  This may require adding some lines
to /etc/system to cement the network interface rate, but this is
not a Hub vs. Switch issue - just a Sun vs. networking issue.
						Ric Anderson [ric@Opus1.COM]
============================================================================
=================
what kind of switches? Cisco, perhaps? autonegotiation between sun and cisco
is HORRIBLE and cannot be trusted. here is a doc that shows how to hard set
the speed and duplex of the ports in question, on the sun side:

http://shells.devunix.org/~argoth/iaoq/
						Mark Luntzel
[mark@neurosis.net]
============================================================================
=================
Shouldn't, just be sure that both switches and hosts are
auto-negotiating at full-duplex 100MB. See the FAQ for
info on this.
						David Foster
[foster@dim.ucsd.edu]
============================================================================
=================
Besides the intrinsic differences in throughput, switches can autonegotiate
the same as NICs; however, you may want to hard set this to guarantee
success and bandwidth capacities.  Secondarily, Fibre Channel switches add
complexity in that you need to be certain that you are aware of Arbitrated
Loop (FC-AL) vs. Switched Fabric (FC-SW) capabilities and
incompatibilities.  Some Fibre Channel switches are FC-SW only while others
can handle FC-SW and FC-AL.  Arbitrated Loop devices to not maintain a true
WWN (World Wide Name), which is in essence an extended MAC address.
	
mjohnson@forsythesolutions.com
============================================================================
=================
The solution is, you can connect your server to the switch withouat any
issues. There is no extra setting or configuration you need to do. You
can try connecting the server to switch, in case of any problems get
back to me.
						Satish Kumar M
[satish.m@sun.com]
============================================================================
=================
Received on Thu Jun 28 21:59:55 2001

This archive was generated by hypermail 2.1.8 : Wed Mar 23 2016 - 16:24:58 EDT