On Fri, Oct 21, 2005 at 12:08:10PM -0400, Harry Tirrell wrote:
> My thanks to all who responded to my question about GigE and autonegotiate on
> AIX, with the consensus being (100%) don't worry about it. My only concern
> is the output of the ifconfig command on that interface which is seen below.
>
> en1:
> flags=4e080863,80<UP,BROADCAST,NOTRAILERS,RUNNING,SIMPLEX,MULTICAST,GROUPRT
> ,64BIT,PSEG,CHAIN>
>
> Am I misinterputing the SIMPLEX
Depends what you think it means, really...
SIMPLEX does NOT mean half-duplex.
The SIMPLEX flag refers to the fact that an Ethernet interface can't
"hear" its own transmissions.
Some OS's report interface media options (including any link status
information) in the rest of the output. Some don't. I'm guessing
AIX doesn't.
> or do I have a network problem, which the network people deny,
I'm not saying that you don't have a network problem.
But why do you think you do? You don't say...
Check your network interface for errors/collisions. Check the switch
port for the same. You should have a tiny number or none. If you
have bad cable, GBICs/patch panels etc. you may have more - but it
may still be acceptable (ish).
If you have huge (and rapidly increasing) numbers for errors/collisions,
runt packets etc. on either end, that's your problem right there.
A crude double-check of interface throughput can be performed by
getting a reasonably powerful host, with a comparable (i.e. GigE)
interface, in the same VLAN. Ideally on the same switch, on the
same blade would be perfect.
Make both hosts relatively idle for the test.
Use a low-level utiiity to blast data over the link. Even FTP is
better than nothing (although it adds more overhead than, say,
netcat). Ponder your figures.
The result should be immediately obvious; any speed or duplex
mismatch will greviously hurt your throughput numbers, and you will
have errors/collisions either on the network interface or on the
switch port. I've seen throughput numbers as bad as 200-300KB/s
for 100BASE-T on an FTP transfer.
I have no idea what gear you're plugged into, but I'd say you should
see at least >80% of rated wire speed, depending on the switch
you're using and the backplane load. Might be an idea to clear this
little test with the netowrk guys first, they tend to get a touch
excitable when you peg the switch CPU or saturate the backplane and
nothing plugged into it has connectivity anymore and all their
beepers fire simulatneously....
Even plugged into a consumer-grade Gig sweitch, you should get
healthy numbers. You won't get anywhere near those figures with a
speed/duplex mismatch.
If you're getting less than rated wire speed but more than 50%, I'd
suspect that either one (or both) of your server(s) simply cannot
drive a GigE pipe, or the switch can't deal. If you suspect this,
simply patch the two servers together and repeat the test - thereby
removing the switch as a variable.
If you're plugged into a $40 switch made in China or Taiwan, well,
uh, "don't do that".
If this works fine but you have issues between blades when you
expand out your test, then the switch backplane might be saturated,
or the Supervisor engine overwhelmed. If the problem lies between
switches, the ISL's might be inadequate etc. etc.
What do your network monitoring tools say?
Cheers,
AS
|
