From: scott (8f27e956gmail.com)
Date: Tue Apr 15 2008 - 01:46:04 CDT

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

It appears that you have two bus types -- PCI-E and PCI-X.

em0 - 5 are PCI-E. PCI-E is a spoke-hub (star) bus topology so each
em() is on its own bus pathway. One PCI-E device does NOT contend with
another.

em6 and em7 are PCI-X and, yes, they're on the same bus, and, yes, they
may contend with each other. Are they, (i) a one dual-ports NIC, or
(ii) two single-port NICs, or (iii) a chip embedded on the mb?

-----Original Message-----
From: Mikael Kermorgant <mikael.kermorgantgmail.com>
Subject: Re: 4.2 and em(4)
Date: Tue, 15 Apr 2008 00:46:08 +0200

Hello,

I'd like to jump on what you said about separate buses because I
haven't looked at this before.

You made me curious to understand this dmesg output :

cpu0 at mainbus0
pci0 at mainbus0 bus 0: configuration mode 1 (no bios)
pchb0 at pci0 dev 0 function 0 "Intel 82Q965 Host" rev 0x02
agp0 at pchb0: aperture at 0xd0000000, size 0x8000000
ppb0 at pci0 dev 1 function 0 "Intel 82Q965 PCIE" rev 0x02: irq 14
pci1 at ppb0 bus 1
ppb1 at pci1 dev 0 function 0 "Intel PCIE-PCIE" rev 0x09
pci2 at ppb1 bus 2
ppb2 at pci1 dev 0 function 2 "Intel PCIE-PCIE" rev 0x09
pci3 at ppb2 bus 3
vga1 at pci0 dev 2 function 0 "Intel 82Q965 Video" rev 0x02
wsdisplay0 at vga1 mux 1: console (80x25, vt100 emulation)
wsdisplay0: screen 1-5 added (80x25, vt100 emulation)
ppb3 at pci0 dev 28 function 0 "Intel 82801H PCIE" rev 0x02: irq 14
pci4 at ppb3 bus 4
em0 at pci4 dev 0 function 0 "Intel PRO/1000MT (82573L)" rev 0x00: irq
14, address 00:10:f3:10:7e:68
ppb4 at pci0 dev 28 function 1 "Intel 82801H PCIE" rev 0x02: irq 10
pci5 at ppb4 bus 5
em1 at pci5 dev 0 function 0 "Intel PRO/1000MT (82573L)" rev 0x00: irq
10, address 00:10:f3:10:7e:69
ppb5 at pci0 dev 28 function 2 "Intel 82801H PCIE" rev 0x02: irq 11
pci6 at ppb5 bus 6
em2 at pci6 dev 0 function 0 "Intel PRO/1000MT (82573L)" rev 0x00: irq
11, address 00:10:f3:10:7e:6a
ppb6 at pci0 dev 28 function 3 "Intel 82801H PCIE" rev 0x02: irq 15
pci7 at ppb6 bus 7
em3 at pci7 dev 0 function 0 "Intel PRO/1000MT (82573L)" rev 0x00: irq
15, address 00:10:f3:10:7e:6b
ppb7 at pci0 dev 28 function 4 "Intel 82801H PCIE" rev 0x02: irq 14
pci8 at ppb7 bus 8
em4 at pci8 dev 0 function 0 "Intel PRO/1000MT (82573L)" rev 0x00: irq
14, address 00:10:f3:10:7e:6c
ppb8 at pci0 dev 28 function 5 "Intel 82801H PCIE" rev 0x02: irq 10
pci9 at ppb8 bus 9
em5 at pci9 dev 0 function 0 "Intel PRO/1000MT (82573L)" rev 0x00: irq
10, address 00:10:f3:10:7e:6d
uhci0 at pci0 dev 29 function 0 "Intel 82801H USB" rev 0x02: irq 5
uhci1 at pci0 dev 29 function 1 "Intel 82801H USB" rev 0x02: irq 15
ehci0 at pci0 dev 29 function 7 "Intel 82801H USB" rev 0x02: irq 5
usb0 at ehci0: USB revision 2.0
uhub0 at usb0 "Intel EHCI root hub" rev 2.00/1.00 addr 1
ppb9 at pci0 dev 30 function 0 "Intel 82801BA Hub-to-PCI" rev 0xf2
pci10 at ppb9 bus 10
em6 at pci10 dev 14 function 0 "Intel PRO/1000MT (82541GI)" rev 0x05:
irq 11, address 00:10:f3:10:7e:6e
em7 at pci10 dev 15 function 0 "Intel PRO/1000MT (82541GI)" rev 0x05:
irq 10, address 00:10:f3:10:7e:6f

Just by reading this :
----
pci4 at ppb3 bus 4
em0 at pci4 dev 0 function 0 "Intel PRO/1000MT (82573L)" rev 0x00: irq
14, address 00:10:f3:10:7e:68
ppb4 at pci0 dev 28 function 1 "Intel 82801H PCIE" rev 0x02: irq 10
pci5 at ppb4 bus 5
em1 at pci5
------
I'd deduce em0 (pci4, bus 4) and em1 (pci5, bus 5) are on separate
buses... but am I right ?
But em6 and em7 are on the same bus, right ?

Thanks in advance,

Mikael Kermorgant

On Mon, Apr 14, 2008 at 11:14 PM, scott <8f27e956gmail.com> wrote:
> We've found the best gateway box -- pf, sshd for "ssh -w" vpn and ipsec
> clients, spamd, etc. -- is non-MP, as follows.
>
> A) Use a box with the fastest memory bandwidth (and latency) your budget
> -- cash or time spent scrounging -- can afford/acquire. (e.g. on a
> P-III 1 GHz machine, we saw meaningful better top-end results on our
> stress tests between using PC133 vs PC100 and again between PC133 CL2.5
> vs CL3 memory sticks.)
>
> B.1) Server-class motherboards usually have multiple PCI buses (say
> again, "buses," not "slots"). Opposing the em(4) nics on separate
> buses, with regard to in-to-out flows, helps quite a bit too. e.g
> internet --- (em0)(bus1)(pf)(bus2)(em1) --- LAN.
>
> B.2) Once a while back, we did see some positive affect by trying to
> share the driver-IRQ for the like em(4). But not too sure about this
> one.
>
> C) We found, on 4.2, if your mb will play nicely, expressly enabling
> ACPI (vs. default APM) functionality seemed to improve the the boxes
> throughput too. In our case, INTEL MOTHERBOARDS. Your mb may not like
> this, though, so use with care and/or wait to 4.3 release.
>
>
>
> -----Original Message-----
> From: Stuart Henderson <stuspacehopper.org>
> To: miscopenbsd.org
> Subject: Re: 4.2 and em(4)
> Date: Mon, 14 Apr 2008 16:23:24 +0000 (UTC)
> Mailer: slrn/0.9.8.1 (OpenBSD)
> Delivered-To: 8f27e956gmail.com
>
> On 2008-04-14, Joe Warren-Meeks <joehole-in-the.net> wrote:
> >
> > If the box was only doing pf stuff, then that would be correct. If you
> > were to put a bunch of ftp-proxys on there too, then MP would help, no?
>
> very little, the bulk data handling is done in kernel by nat/rdr
> rules added to the anchors, ftp-proxy only touches the control
> connections.

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]