Re: Tuning for pppoe over fibre 30M/1M link

2012-06-02 Thread Stuart Henderson
On 2012-05-29, David Diggles da...@elven.com.au wrote:
 I would love to get 3MB/s, but maybe 1.8MB/s is the limit of the
 realtek NIC.

There are various different realtek-based nics. rl(4) are generally quite
poor and need a fair bit of CPU power to drive. re(4) should be a bit
better. But Geode is slow anyway especially for I/O, and 300MHz is not
a particularly fast Geode. 1.8MB/s (~14Mb/s) seems a bit slow but not
out of the realms of possibility especially if it's rl not re.

 I have just ordered an Atom 1.8GHz with Gigabit Intel NICs, should
 be more than good enough as an upgrade?  I may upgrade my link from
 30Mbit to 100Mbit in future, I would expect the Atom to handle this.

Should be a fair bit better.



Re: Tuning for pppoe over fibre 30M/1M link

2012-05-31 Thread David Diggles
FYI I have now run the same pppoe(4) download test on core2duo with OpenBSD 5.1,
on em0 interface.  It beats the Mac.

Mac G5 dual core 2GHz
3MB/s
Intel core2duo 3GHz OpenBSD i386
3.44MB/s

I have found on the Geode 300MHz, cleaning up the pf.conf, removing
modulate state, and no-df from scrub improves the throughput to:
1.8MB/s.

I ordered an Atom with 3 Intel NICs, just arrived!

Looking forward to testing it out.

The test I am doing is

ftp -o /dev/null http://mirror.internode.on.net/pub/OpenBSD/5.1/src.tar.gz

Internode is my ISP.

On Mon, May 28, 2012 at 05:07:01PM +0200, Andre Keller wrote:
 Am 28.05.2012 15:26, schrieb David Diggles:
  Maybe I should try some of the kernel tuning suggested on calomel.
 
 I would not even visit that site... It's mostly a waste of time as most
 of the tunings are not up-to-date or just plain wrong. OpenBSD ships
 with pretty sane defaults that normally do not need any tweaking unless
 you run some unorthodox configuration. If you need to tweak something
 look into the faq and the sysctl(3) man page and not to calomel.org
 
 
 Could you please be a bit more specific about your setup?
 
 Are you using pppoe(4) or pppoe(8)?
 
 Do you see maxed out mbufs (netstat -m), a very high interrupt load (top
 / vmstat -i), ifq drops (sysctl net.inet.ip.ifq.drops), interface errors
 (netstat -i)?
 
 I'm running pppoe(4) on a lot of Geode 500MHz powered boxes and have no
 problem getting 30Mbit/s throughput of unencrypted traffic...
 
 
 
 g
 Andri



Re: Tuning for pppoe over fibre 30M/1M link

2012-05-31 Thread Bentley, Dain
You'll find the atom performance is good. I have an atom 330 with 2 gigs of
RAM and dual on board NICs with another PCI NIC and it runs great. I also have
openvpn in bridge mode and squid running and no hiccups ar all.

Regards,
Dain Bentley

-Original Message-
From: David Diggles [da...@elven.com.au]
Received: Thursday, 31 May 2012, 4:51pm
To: misc@openbsd.org [misc@openbsd.org]
Subject: Re: Tuning for pppoe over fibre 30M/1M link

FYI I have now run the same pppoe(4) download test on core2duo with OpenBSD
5.1,
on em0 interface.  It beats the Mac.

Mac G5 dual core 2GHz
3MB/s
Intel core2duo 3GHz OpenBSD i386
3.44MB/s

I have found on the Geode 300MHz, cleaning up the pf.conf, removing
modulate state, and no-df from scrub improves the throughput to:
1.8MB/s.

I ordered an Atom with 3 Intel NICs, just arrived!

Looking forward to testing it out.

The test I am doing is

ftp -o /dev/null http://mirror.internode.on.net/pub/OpenBSD/5.1/src.tar.gz

Internode is my ISP.

On Mon, May 28, 2012 at 05:07:01PM +0200, Andre Keller wrote:
 Am 28.05.2012 15:26, schrieb David Diggles:
  Maybe I should try some of the kernel tuning suggested on calomel.

 I would not even visit that site... It's mostly a waste of time as most
 of the tunings are not up-to-date or just plain wrong. OpenBSD ships
 with pretty sane defaults that normally do not need any tweaking unless
 you run some unorthodox configuration. If you need to tweak something
 look into the faq and the sysctl(3) man page and not to calomel.org


 Could you please be a bit more specific about your setup?

 Are you using pppoe(4) or pppoe(8)?

 Do you see maxed out mbufs (netstat -m), a very high interrupt load (top
 / vmstat -i), ifq drops (sysctl net.inet.ip.ifq.drops), interface errors
 (netstat -i)?

 I'm running pppoe(4) on a lot of Geode 500MHz powered boxes and have no
 problem getting 30Mbit/s throughput of unencrypted traffic...



 g
 Andri



Re: Tuning for pppoe over fibre 30M/1M link

2012-05-30 Thread David Diggles
On Tue, May 29, 2012 at 07:23:32PM +1000, David Diggles wrote:
[ snip ]
 http://bincrow.net/test.log
[ snip ]

Interesting, this single post got http://bincrow.net added to the Websense 
blocklist.

Category:

This Websense category is filtered: Potentially Damaging Content. Sites in 
this category
may pose a security threat to the Departments network and are blocked as per 
the Departments
'Use of Internet, Email  Other ICT Facilities  Devices' policy.


All it serves is an index.html, basic html no javascript, and the log I posted.

I guess this list gets trawled for bad urls by content filtering providers.



Re: Tuning for pppoe over fibre 30M/1M link

2012-05-29 Thread David Diggles
Andre, as promised;

Here are the outputs you have asked for, but on the Geode 300MHz.

Throughputs, http downloading src.tar.gz from my ISP mirror in a loop:

Tue May 29 16:33:45 EST 2012 1.84 MB/s
Tue May 29 16:35:01 EST 2012 1.86 MB/s
Tue May 29 16:36:17 EST 2012 1.87 MB/s

The same test when I do pppoe on the Mac gets 3 MB/s.

Outputs of various stat commands while this was happening:

http://bincrow.net/test.log

Note: net.inet.ip.ifq.drops=193 does not change.  I think this was from
when I unplugged the cable earlier.

.d.d.

On Tue, May 29, 2012 at 03:12:03PM +1000, David Diggles wrote:
  Could you please be a bit more specific about your setup?
 
 Sure.
 
  Are you using pppoe(4) or pppoe(8)?
 
 pppoe(4)
 
  Do you see maxed out mbufs (netstat -m), a very high interrupt load (top
  / vmstat -i), ifq drops (sysctl net.inet.ip.ifq.drops), interface errors
  (netstat -i)?
 
 None of the above were maxed out on the P4.  It was only a quick test,
 as this is the production spamd server.
 
  I'm running pppoe(4) on a lot of Geode 500MHz powered boxes and have no
  problem getting 30Mbit/s throughput of unencrypted traffic...
 
 I plugged it back into the gw, Geode 300MHz with 100MBit Realtek.
 
 I made the pf.conf as default as possible (to look like the
 example pf.conf provided in /etc), I removed all the modulate and
 synproxy state options that calomel suggested putting in pf.conf.
 
 The performance improved from 1MB/s to 1.8MB/s.
 
 I would love to get 3MB/s, but maybe 1.8MB/s is the limit of the
 realtek NIC.
 
 I have just ordered an Atom 1.8GHz with Gigabit Intel NICs, should
 be more than good enough as an upgrade?  I may upgrade my link from
 30Mbit to 100Mbit in future, I would expect the Atom to handle this.
 
 .d.d.



Re: Tuning for pppoe over fibre 30M/1M link

2012-05-28 Thread David Diggles
I have got it to do 10Mbps now, by ditching the 85Mbps ethernet over
power adaptors, in favor of a cable.

I get 12Mbps if I run it to the 2.4GHz Pentium 4 xl0 100Mbps port.

No idea what is slowing it down here yet.  It should be getting 30Mbps,
like it does on the Mac.

Maybe I should try some of the kernel tuning suggested on calomel.

On Mon, May 21, 2012 at 11:00:22AM -0600, Daniel Melameth wrote:
 On Mon, May 21, 2012 at 9:35 AM, David Diggles da...@elven.com.au wrote:
  I am still getting 300 kilobytes/second download speed with OpenBSD pppoe, 
  however when
  I plug directly into a Mac and run pppoe on it, 3 megabytes/second.
 
  What should I look at for tuning this to get 3MB/s through OpenBSD?
 
  Connection: pppoe, over fibre, 30M downlink, 1M uplink
 
  The OpenBSD gateway is using the kernel pppoe driver.
  ...
  OpenBSD 5.1 (GENERIC) #160: Sun Feb 12 09:46:33 MST 2012
  ? ?dera...@i386.openbsd.org:/usr/src/sys/arch/i386/compile/GENERIC
  cpu0: Geode(TM) Integrated Processor by National Semi (CyrixInstead 
  586-class) 301 MHz
  cpu0: FPU,TSC,MSR,CX8,CMOV,MMX
  real mem ?= 132182016 (126MB)
  avail mem = 119992320 (114MB)
 ...
  rl0 at pci0 dev 14 function 0 Realtek 8139 rev 0x10: irq 12, address 
  00:90:0b:04:bb:f1
  rlphy0 at rl0 phy 0: RTL internal PHY
  rl1 at pci0 dev 15 function 0 Realtek 8139 rev 0x10: irq 10, address 
  00:90:0b:04:bb:f2
  rlphy1 at rl1 phy 0: RTL internal PHY
  rl2 at pci0 dev 16 function 0 Realtek 8139 rev 0x10: irq 11, address 
  00:90:0b:04:bb:f3
  rlphy2 at rl2 phy 0: RTL internal PHY
 
 FWIW, I have 20M/5M VDSL service at home and have zero issue doing
 20Mbps with OpenBSD as my pppoe-based firewall.  That said, while I
 wouldn't expect a 300MHz machine to limit you to 2.4Mbps, it is a bit
 weak--and rl NICs are some of the worst out there.  Curiously, when
 doing 2.4Mbps, what does top show for interrupts?  For comparison,
 when I'm doing 20Mbps, my interrupts are at 5-6% using em and fxp
 NICs.



Re: Tuning for pppoe over fibre 30M/1M link

2012-05-28 Thread Andre Keller
Am 28.05.2012 15:26, schrieb David Diggles:
 Maybe I should try some of the kernel tuning suggested on calomel.

I would not even visit that site... It's mostly a waste of time as most
of the tunings are not up-to-date or just plain wrong. OpenBSD ships
with pretty sane defaults that normally do not need any tweaking unless
you run some unorthodox configuration. If you need to tweak something
look into the faq and the sysctl(3) man page and not to calomel.org


Could you please be a bit more specific about your setup?

Are you using pppoe(4) or pppoe(8)?

Do you see maxed out mbufs (netstat -m), a very high interrupt load (top
/ vmstat -i), ifq drops (sysctl net.inet.ip.ifq.drops), interface errors
(netstat -i)?

I'm running pppoe(4) on a lot of Geode 500MHz powered boxes and have no
problem getting 30Mbit/s throughput of unencrypted traffic...



g
Andri



Re: Tuning for pppoe over fibre 30M/1M link

2012-05-28 Thread David Diggles
 Could you please be a bit more specific about your setup?

Sure.

 Are you using pppoe(4) or pppoe(8)?

pppoe(4)

 Do you see maxed out mbufs (netstat -m), a very high interrupt load (top
 / vmstat -i), ifq drops (sysctl net.inet.ip.ifq.drops), interface errors
 (netstat -i)?

None of the above were maxed out on the P4.  It was only a quick test,
as this is the production spamd server.

 I'm running pppoe(4) on a lot of Geode 500MHz powered boxes and have no
 problem getting 30Mbit/s throughput of unencrypted traffic...

I plugged it back into the gw, Geode 300MHz with 100MBit Realtek.

I made the pf.conf as default as possible (to look like the
example pf.conf provided in /etc), I removed all the modulate and
synproxy state options that calomel suggested putting in pf.conf.

The performance improved from 1MB/s to 1.8MB/s.

I would love to get 3MB/s, but maybe 1.8MB/s is the limit of the
realtek NIC.

I have just ordered an Atom 1.8GHz with Gigabit Intel NICs, should
be more than good enough as an upgrade?  I may upgrade my link from
30Mbit to 100Mbit in future, I would expect the Atom to handle this.

.d.d.



Tuning for pppoe over fibre 30M/1M link

2012-05-21 Thread David Diggles
Now I'm all upgraded to 5.1 I'm very happy with it all, other than a few 
minor issues,
most notably:

I am still getting 300 kilobytes/second download speed with OpenBSD pppoe, 
however when
I plug directly into a Mac and run pppoe on it, 3 megabytes/second.

What should I look at for tuning this to get 3MB/s through OpenBSD?

Connection: pppoe, over fibre, 30M downlink, 1M uplink

The OpenBSD gateway is using the kernel pppoe driver.
...
OpenBSD 5.1 (GENERIC) #160: Sun Feb 12 09:46:33 MST 2012
dera...@i386.openbsd.org:/usr/src/sys/arch/i386/compile/GENERIC
cpu0: Geode(TM) Integrated Processor by National Semi (CyrixInstead 
586-class) 301 MHz
cpu0: FPU,TSC,MSR,CX8,CMOV,MMX
real mem  = 132182016 (126MB)
avail mem = 119992320 (114MB)
mainbus0 at root
bios0 at mainbus0: AT/286+ BIOS, date 07/18/01, BIOS32 rev. 0 @ 0xfae40, SMBIOS 
rev. 2.2 @ 0xf0800 (42 entries)
bios0: vendor Award Software International, Inc. version 6.00 PG date 
07/18/2001
apm0 at bios0: Power Management spec V1.2
acpi at bios0 function 0x0 not configured
pcibios0 at bios0: rev 2.1 @ 0xf/0xb2c8
pcibios0: PCI IRQ Routing Table rev 1.0 @ 0xfdb70/80 (3 entries)
pcibios0: PCI Exclusive IRQs: 10 11 12
pcibios0: no compatible PCI ICU found: ICU vendor 0x1078 product 0x0100
pcibios0: Warning, unable to fix up PCI interrupt routing
pcibios0: PCI bus #0 is the last bus
bios0: ROM list: 0xc/0x8000
cpu0 at mainbus0: (uniprocessor)
cpu0: TSC disabled
pci0 at mainbus0 bus 0: configuration mode 1 (bios)
pchb0 at pci0 dev 0 function 0 Cyrix GXm PCI rev 0x00
rl0 at pci0 dev 14 function 0 Realtek 8139 rev 0x10: irq 12, address 
00:90:0b:04:bb:f1
rlphy0 at rl0 phy 0: RTL internal PHY
rl1 at pci0 dev 15 function 0 Realtek 8139 rev 0x10: irq 10, address 
00:90:0b:04:bb:f2
rlphy1 at rl1 phy 0: RTL internal PHY
rl2 at pci0 dev 16 function 0 Realtek 8139 rev 0x10: irq 11, address 
00:90:0b:04:bb:f3
rlphy2 at rl2 phy 0: RTL internal PHY



Re: Tuning for pppoe over fibre 30M/1M link

2012-05-21 Thread David Diggles
 FWIW, I have 20M/5M VDSL service at home and have zero issue doing
 20Mbps with OpenBSD as my pppoe-based firewall.  That said, while I
 wouldn't expect a 300MHz machine to limit you to 2.4Mbps, it is a bit
 weak--and rl NICs are some of the worst out there.  Curiously, when
 doing 2.4Mbps, what does top show for interrupts?  For comparison,
 when I'm doing 20Mbps, my interrupts are at 5-6% using em and fxp
 NICs.

Good point.  1MB/s is the best I can get with this HW on a local
network test.  The interrupts max out.

I would still like to find out why I only get 300KB/s through the pppoe.
1MB/s would be a nice improvement, and I could put up with it, for
sake of having OpenBSD security until I get around to upgrading the HW.

Can anyone recommend an embedded product that would be adequette for
100Mbps? (thinking of future network upgrades).

Cheers,
.d.d.