> Why not a good Intel card?

I thought D-Link would be acceptable.. and the shop didnt have Intel cards..

>> Btw. what an earth is MII PHY?
>
> MII is the Media Independent Interface (IEE 802.3), and is a (serial) bus
> that is used to connect MACs to PHYs (physical media interfaces).

Thank you!

On Tue, 31 May 2005 12:32:15 +0300, Mr.Jason wrote:

>> I have had no problems when using the DLINK card. The card
>> uses the 8139too module.
>>
>> Here is the relevent info from lspci:
>> 00:09.0 Ethernet controller: D-Link System Inc RTL8139 Ethernet (rev 10)
>>
>> And from dmesg:
>> eth0: RealTek RTL8139 at 0xe09c3000, 00:50:xx:xx:xx:xx, IRQ 11
>> eth0: Identified 8139 chip type 'RTL-8139C'
>
> Thank you! Do you think that card is good enough for server? I mean
> stability issue, it wont go down all of a sudden? Or if theres lots of
> traffic or even DOS attack..
> Mr.Jason
> http://music.msn.com/album/?album=41839578
>
These inexpensive cards work for me. YMMV.

I have a couple of simple firewall/routers which are dual NIC with these
cards. The cards don't seem to be the critical mode of failure in these
boxes (but come to think of it, these boxes work great all the time.)
None of my equipment is heavily loaded. YMMV.

There is another box with dual Netgear FA310 (another inexpensive card)
and it works fine also. Both the Netgear card and the DLink card are
supported for use with SMP hardware/kernels. YMMV.

On Tue, 31 May 2005 15:46:14 +0200, Daniel de Kok <daniel@mindbender.nowhere> wrote:
>> eth0: VIA VT6105 Rhine-III
>> eth0: MII PHY
>
>Why not a good Intel card?

Maybe comes with Via chipset, on mobo? Waits for 'sempro' power up...

00:12.0 Ethernet controller: VIA Technologies, Inc. VT6102 [Rhine-II] (rev 78)
This one is on mobo in 'sempro' box. winxp (pro/100) --> to slack on via
Rhine-II 85% link saturation on large single file transfer (samba), usually
around 50% to other boxen.

In firewall box, for localnet, lspci:
00:0d.0 Ethernet controller: Intel Corp. 82557/8/9 [Ethernet Pro 100] (rev 08)

modules.conf:
# 192.168.1.0/24
alias eth0 e100
# 192.168.2.0/24
#alias eth1 de4x5
# ADSL modem
alias eth1 eepro
options eepro io=0x280 irq=3
#

Old ISA Intel for modem half-duplex link. I used to run second localnet
segment with DE450 combo NIC, they pretty good too (but _not_ with tulip
driver). Disable COM2 in mobo BIOS to free up IRQ3 with older boxen.

Compare Intel & Realtek (another box has one of each):

00:0d.0 Ethernet controller: Intel Corporation 82557/8/9 [Ethernet Pro 100] (rev 08)
Subsystem: Intel Corporation EtherExpress PRO/100+ Management Adapter
Flags: bus master, medium devsel, latency 66, IRQ 11
Memory at f4100000 (32-bit, non-prefetchable) [size=4K]
I/O ports at 1000 [size=64]
Memory at f4000000 (32-bit, non-prefetchable) [size=1M]
Capabilities: [dc] Power Management version 2

00:0f.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL-8139/8139C/8139C+ (rev 10)
Subsystem: AOPEN Inc.: Unknown device 0027
Flags: bus master, medium devsel, latency 64, IRQ 10
I/O ports at 1400 [size=256]
Memory at f4101000 (32-bit, non-prefetchable) [size=256]
Capabilities: [50] Power Management version 2

The Intel has far larger interface memory 'footprint'. OS transfers
whole packets to/from it, whereas RTL-8139 transfers packets in little
pieces. The ubiquitous RTL-8139c probably gets a bad rep. due to poor
layouts on cheapie NICs, wrong PCI timing issues for MMIO, not so much
the chip itself.

I've not had 'lockup' problems with _quality_ RTL-8139c based NICs.

I used to accept 50% 'network utilization' displayed by winxp task
manager on a large file write (samba) to 'peetoo' (pII/350) with 8139c.
Dropped a 700MB file on link with pro/100, and I get 75% - 78% now.

>MII is the Media Independent Interface (IEE 802.3), and is a (serial) bus
>that is used to connect MACs to PHYs (physical media interfaces).

Always trips me some when I read PHY, abbreviation rather than acronym.

--Grant.

On Tue, 31 May 2005 23:11:24 +0000, Grant Coady wrote:
> Maybe comes with Via chipset, on mobo? Waits for 'sempro' power up...

Mr. Jason bought a new D-Link card.

> The Intel has far larger interface memory 'footprint'. OS transfers
> whole packets to/from it, whereas RTL-8139 transfers packets in little
> pieces. The ubiquitous RTL-8139c probably gets a bad rep. due to poor
> layouts on cheapie NICs, wrong PCI timing issues for MMIO, not so much
> the chip itself.

Wrong, it does because of the chip itself. The NIC chip does not support
many important performace features that offload CPU usage[1], besides that
the design is essentially flawed[2].

[1] http://www.fefe.de/linuxeth/
[2] http://www.fefe.de/linuxeth/realtek.txt

-- Daniel

On Wed, 01 Jun 2005 07:43:19 +0200, Daniel de Kok <daniel@mindbender.nowhere> wrote:
>
>Wrong, it does because of the chip itself. The NIC chip does not support
>many important performace features that offload CPU usage[1], besides that
>the design is essentially flawed[2].
>
>[1] http://www.fefe.de/linuxeth/
>[2] http://www.fefe.de/linuxeth/realtek.txt

That is not evidence, just an anonymous anecdote, you've done no
research at all, have you?

RTL-8139 != RTL-8139C or later.

--Grant.

> Mr. Jason bought a new D-Link card.
>
>> The Intel has far larger interface memory 'footprint'. OS transfers
>> whole packets to/from it, whereas RTL-8139 transfers packets in little
>> pieces. The ubiquitous RTL-8139c probably gets a bad rep. due to poor
>> layouts on cheapie NICs, wrong PCI timing issues for MMIO, not so much
>> the chip itself.
>
> Wrong, it does because of the chip itself. The NIC chip does not support
> many important performace features that offload CPU usage[1], besides that
> the design is essentially flawed[2].

>
> [1] http://www.fefe.de/linuxeth/
> [2] http://www.fefe.de/linuxeth/realtek.txt
>
> -- Daniel

Thank you for useful information. Glad I got rid of my RTL.
(BTW. Anyone wanna buy (little) used SMC Ethernet Card? ) haha only kidding.

> RTL-8139 != RTL-8139C or later.
>
> --Grant.
>

Miner was C and it did two hangups (eth0 went down) because of high traffic.
Only reboot helped.
Now I just wait and see what comes up with the new card. Hopefully nothing.

As its been said, "No news is best news"

On Wed, 01 Jun 2005 06:58:40 +0000, Grant Coady wrote:
> That is not evidence, just an anonymous anecdote,

For your interest, the note in the second URL is from the FreeBSD driver
source. Tell me who whe should rather believe concerning the quality of
Realtek 8139-based cards. You, or the person who wrote the Realtek driver
for FreeBSD?

> RTL-8139 != RTL-8139C or later.

Have you read the driver sources? It does not make a real difference
whether you have a RTL8139, RTL8139B, RTL8139C or RTL8139D. The 8139
realtek chips want longword aligned TX buffers, and there can only be one
fragment buffer per packet. Besides that most drivers (e.g. in Linux and
FreeBSD) disable use of MMIO, because it is to broken to function
correctly on SMP machines.

-- Daniel

On Wed, 01 Jun 2005 10:54:23 +0200, Daniel de Kok wrote:
> FreeBSD) disable use of MMIO, because it is to broken to function

s/to/too/

On Wed, 01 Jun 2005 10:54:23 +0200, Daniel de Kok <daniel@mindbender.nowhere> wrote:
> Tell me who whe should rather believe concerning the quality of
>Realtek 8139-based cards. You, or the person who wrote the Realtek driver
>for FreeBSD?

Ask someone who cares, technology is not a belief system, I choose
Intel pro/100 over RTL-8139 for a 50% measured performance boost on
_my_ equipment, and I trust the person taking the measurements.

I don't give a rat's ring for your opinion, I'm secure in my
competency. Opinions are like arseholes, everybody has one.

--Grant.