Re: [pfSense Support] iperf question

2005-08-20 Thread Randy B

Fleming, John (ZeroChaos) wrote:

I'd also like to know which rl cards these are. Can you send the output
of pciconf -lv?


Glad to oblige

[EMAIL PROTECTED]:9:0:   class=0x02 card=0x13011186 chip=0x13001186 rev=0x10 
hdr=0x00

vendor   = 'D-Link System Inc'
device   = 'DL 10038C or 10038D (Remark of Realtek RTL-8139) Fast 
Ethernet Adapter'

class= network
subclass = ethernet
[EMAIL PROTECTED]:10:0: class=0x02 card=0xf3111385 chip=0x0020100b rev=0x00 
hdr=0x00

vendor   = 'National Semiconductor'
device   = 'DP83815/16 Fast Ethernet Adapter (MacPhyter/MacPhyter-II)'
class= network
subclass = ethernet
[EMAIL PROTECTED]:11:0:  class=0x02 card=0x13011186 chip=0x13001186 rev=0x10 
hdr=0x00

vendor   = 'D-Link System Inc'
device   = 'DL 10038C or 10038D (Remark of Realtek RTL-8139) Fast 
Ethernet Adapter'

class= network
subclass = ethernet


Chris Buechler wrote:
  Yes it is.  iperf doesn't test full duplex, it's one direction only

(with one connection, run a server and a client on each side and you
can test full duplex).  You'll never get more than 100 Mb on a 100Mb
link or 10 Mb on a 10 Mb link, even if it's full duplex, with a single
iperf server and client.

The specific command I ran was iperf -i 1 -N -d -P3 -c 192.168.0.1 - 
from the options on my Gentoo box, -d says it does a bidirectional test 
simultaneously, testing (I presumed) duplex.



rl's are known for poor performance, but should be better than that
unless you're only running a 100-200 MHz machine or so.


I just barely miss that category... ;-)
CPU: AMD-K6(tm) 3D processor (300.68-MHz 586-class CPU)


You should be seeing:
media: Ethernet autoselect (100baseTX full-duplex) 
in your ifconfig output.  Exactly what are you seeing on that line?


rl0: flags=8843UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST mtu 1500
options=8VLAN_MTU
inet 192.168.0.1 netmask 0xff00 broadcast 192.168.0.255
inet6 fe80::211:95ff:fe28:ab2f%rl0 prefixlen 64 scopeid 0x1
ether 00:11:95:28:ab:2f
media: Ethernet autoselect (100baseTX full-duplex)
status: active


-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



Re: [pfSense Support] iperf question

2005-08-20 Thread Chris Buechler
 The specific command I ran was iperf -i 1 -N -d -P3 -c 192.168.0.1 -
 from the options on my Gentoo box, -d says it does a bidirectional test
 simultaneously, testing (I presumed) duplex.
 

ah yeah, it is full duplex with that option.  I assumed you were doing
nothing but a -c and -s.


  rl's are known for poor performance, but should be better than that
  unless you're only running a 100-200 MHz machine or so.
 
 I just barely miss that category... ;-)
 CPU: AMD-K6(tm) 3D processor (300.68-MHz 586-class CPU)
 

hah  Well...that's probably the best you can get on that.  :)  With rl
NIC's at least, since they're interrupt happy.  When you're testing
throughput, can you try to run 'top' at the console or a SSH session? 
I'm curious what your CPU utilization will be.

I had a rl NIC in a P3 600 FreeBSD box, and it could only do about 70
Mb to another host on my LAN.  Put a Intel fxp in the same box, and it
could do 100 Mb at wire speed.  With an Intel gig 'em' card, the same
box can do 400 Mb though a single NIC.  Considering that when you're
passing traffic, you can roughly cut that number in half, that P3 600
could have only done probably 35 Mb in a firewalling scenario with rl
NIC's.  Yes, they really are that bad.  :)  At 70 Mb with the rl, the
P3 600 was pegged at 100% CPU, mostly from interrupts.

A P3 600 is easily 2-3 times as fast as a K6 300, so those numbers
don't look too out of wack.


 rl0: flags=8843UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST mtu 1500
 media: Ethernet autoselect (100baseTX full-duplex)
 

looks fine.  I bet if you replace the rl NIC's with fxp's, you'll see
a huge improvement in performance.

-cmb

-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



Re: [pfSense Support] iperf question

2005-08-20 Thread Randy B

Chris Buechler wrote:

hah  Well...that's probably the best you can get on that.  :)  With rl
NIC's at least, since they're interrupt happy.


Wow.  That was certainly it.  Ran top and showed 0% idle CPU with over 
70% interrupt dedicated to interrupts and ~25% system.  I knew the RL 
NICs were poor, just never knew how poor they really were until I 
started playing around with BSD - I guess my Linux machines have always 
been powerful enough to overcome the danged things.  Funny this - the 
93Mb was between a desktop Athlon XP-1800 and a laptop AMD-64 3000+, 
both with RTL-8139 NICs.


I guess I'll stop buying the crappy RTL cards now, eh?

Hey, anyone interested in a couple of top-quality NICs?  I'll sell 'em 
to you cheap!


RB


-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



Re: [pfSense Support] iperf question

2005-08-19 Thread Chris Buechler
On 8/18/05, Randy B [EMAIL PROTECTED] wrote:
 Chris Buechler wrote:
  Not unless you're running both a client and server at each end.
 
 Unfortunately, not the case - 

Yes it is.  iperf doesn't test full duplex, it's one direction only
(with one connection, run a server and a client on each side and you
can test full duplex).  You'll never get more than 100 Mb on a 100Mb
link or 10 Mb on a 10 Mb link, even if it's full duplex, with a single
iperf server and client.



 
 I'm able to get 93Mb to another machine on the network - acceptable,
 given the cheap switch I have.  
 

that's roughly as good as you're going to get on 100 Mb.  


 
 I have two rl cards and one sis - sis0 is linked to my cable modem and
 my LAN is to rl0.  The RL NICs are both rather new, and both say they've
 autonegotiated at 100Mb.  

rl's are known for poor performance, but should be better than that
unless you're only running a 100-200 MHz machine or so.

what duplex does it say?  or does it not say?  I'm still thinking
duplex mismatch, though 20 something Mb is quite a bit for having a
mismatch.

You should be seeing:
media: Ethernet autoselect (100baseTX full-duplex)

in your ifconfig output.  Exactly what are you seeing on that line?

-cmb

-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



RE: [pfSense Support] iperf question

2005-08-19 Thread Fleming, John \(ZeroChaos\)
I'd also like to know which rl cards these are. Can you send the output
of pciconf -lv?

thanks

-Original Message-
From: Chris Buechler [mailto:[EMAIL PROTECTED] 
Sent: Friday, August 19, 2005 8:31 AM
Cc: support@pfsense.com
Subject: Re: [pfSense Support] iperf question

On 8/18/05, Randy B [EMAIL PROTECTED] wrote:
 Chris Buechler wrote:
  Not unless you're running both a client and server at each end.
 
 Unfortunately, not the case - 

Yes it is.  iperf doesn't test full duplex, it's one direction only
(with one connection, run a server and a client on each side and you
can test full duplex).  You'll never get more than 100 Mb on a 100Mb
link or 10 Mb on a 10 Mb link, even if it's full duplex, with a single
iperf server and client.



 
 I'm able to get 93Mb to another machine on the network - acceptable,
 given the cheap switch I have.  
 

that's roughly as good as you're going to get on 100 Mb.  


 
 I have two rl cards and one sis - sis0 is linked to my cable modem and
 my LAN is to rl0.  The RL NICs are both rather new, and both say
they've
 autonegotiated at 100Mb.  

rl's are known for poor performance, but should be better than that
unless you're only running a 100-200 MHz machine or so.

what duplex does it say?  or does it not say?  I'm still thinking
duplex mismatch, though 20 something Mb is quite a bit for having a
mismatch.

You should be seeing:
media: Ethernet autoselect (100baseTX full-duplex)

in your ifconfig output.  Exactly what are you seeing on that line?

-cmb

-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]


-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



AW: [pfSense Support] iperf question

2005-08-18 Thread Holger Bauer
From the m0n0 documentation (hidden options in the config.xml file). Same like 
m0n0:

interfaces/(if)/media and interfaces/(if)/mediaopt

If you need to force your NIC to a specific media type (e.g. 10Base-T half 
duplex), you can use these two options. Refer to the appropriate FreeBSD 
manpage for the driver you're using to see which options are available (or run 
ifconfig -m).

You can diagnosticsedit file /conf/config.xml to make this from the webgui.

Example (but issue ifconfig -m from diagnosticscommand to see the options of 
your specific card first, they might be different):
opt1
descrMichelbach entfernt/descr
ifsis2/if
mtu576/mtu
media10baseT/UTP/media
mediaoptfull-duplex/mediaopt
ipaddr10.10.15.10/ipaddr
subnet24/subnet
bridge/
enable/
/opt1

Holger

-Ursprüngliche Nachricht-
Von: Randy B [mailto:[EMAIL PROTECTED]
Gesendet: Donnerstag, 18. August 2005 04:51
An: support@pfsense.com
Betreff: [pfSense Support] iperf question


I know this isn't likely the best forum for this question, but please 
bear with me.

I've been seeing a lot of these iperf comments/questions, and decided to 
try to track down why my connection to my home firewall seems *so slow*. 
  Installed the package and fired it up, and sure enough - although both 
ends are reporting they're negotiated at 100Mb/s, I'm only getting 
~22Mb, which reeks strongly of 10Mb full-duplex.  It's 
switch-independent - already swapped that out and tried.  My next step 
is to hook a laptop directly up to both machines and give it a whirl 
(rule out a faulty cable).

In the meantime, is there anything I can do on pfSense to fiddle with 
autonegotiation settings like I can with ethtool on my Linux machines? 
I don't have any traffic shaping set up, so I can't see why that would 
come into play, but I'm all ears here.

RB

*BTW - I noticed a while back that pfSense had my favorite alias 'll'. 
It doesn't now, and looking in root's home directory, it looks like 
.tcshrc is zeroed out - either it was originally linked to .cshrc or 
someone's whacked it.  Minor bug that should be ironed out before beta, 
I should think - I'll post a ticket when I get around to it, unless 
someone beats me to it (fixing or otherwise).

-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



Virus checked by G DATA AntiVirusKit


-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]



Re: [pfSense Support] iperf question

2005-08-18 Thread Chris Buechler
On 8/17/05, Randy B [EMAIL PROTECTED] wrote:
 I know this isn't likely the best forum for this question, but please
 bear with me.
 
 I've been seeing a lot of these iperf comments/questions, and decided to
 try to track down why my connection to my home firewall seems *so slow*.
  Installed the package and fired it up, and sure enough - although both
 ends are reporting they're negotiated at 100Mb/s, I'm only getting
 ~22Mb, which reeks strongly of 10Mb full-duplex.  

Not unless you're running both a client and server at each end.  iperf
is only send or receive (by default).  Even at that, you'd max out at
20 Mb.  If your ifconfig says it's 100 Mb, it's 100 Mb.

I'd look elsewhere.  Messing with speed and duplex is more likely to
cause problems (duplex mismatch) than anything.  Unless you have a
duplex mismatch already, are both ends negotiating the same duplex? 
If so, what are the specs of your hardware?

-cmb

-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]