Ethernet High-Performance LAN Adapter (FC #2980)
UZnal
8EF5 (2-1) Ethernet High-Performance LAN Adapter (FC #2980)
I am looking for a layout or a picture of this card showing every detail.
William R. Walsh
> 8EF5 (2-1) Ethernet High-Performance LAN Adapter (FC #2980)
> I am looking for a layout or a picture of this card showing every detail.
The 3C527 and this card are the same thing with different card-IDs.
I have both versions, although none of mine have that long metallic
handle on the back. If you want one, post here or send mail to wct
<atsign> walshcomptech <dot> com.
I can also scan one if you think that will result in a good enough
result.
William
UZnal
> The 3C527 and this card are the same thing with different card-IDs.
Now I know where I have seen it, I should have a 3C527. But then, the type
2-1 card could be assigned to 0041.ADF renamed to 8EF5.ADF, and the drivers,
provided they do not check the adapter ID, work for the type 2-1 card (if
not, the adapter ID can be patched in the driver object code). I saw some of
the 8EF5 driver code, the POS bytes match the 0041.ADF quite exactly.
ADF 0041 3Com EtherLink/MC 32 Ethernet Adapter (3C527) == 8EF5
2-1 8EF5 Ethernet High-Performance LAN Adapter [ent] == 0041
9-Q 8EF5 10/100 Mbps Ethernet UNI High-Performance Ethernet Adapter == ?
9-K 8F62 10/100 Mbps Ethernet MC Adapter (SMP and UNI) [srent, San Remo]
9-Q and 2-1 share the same adapter ID 8EF5, does that mean they share the
same AIX driver? Louis claims in the Tool that 9-Q and 9-K are identical,
was the adapter ID 8F62 perhaps introduced to differentiate between the UNI
and SMP versions or does it mean more?
In my RS/6000 adapter list, 9-K (alias "San Remo" for remo-delled?) has two
feature codes and contains the feature code of 9-Q:
2-1 8EF5 Ethernet High-Performance LAN Adapter (FC#2980)
9-Q 8EF5 10/100 Mbps Ethernet UNI High-Performance Ethernet (FC#2964)
9-K 8F62 10/100 Mbps Ethernet MC Adapter (SMP and UNI) (FC#2964 #2994)
RickE
> 8EF5 (2-1) Ethernet High-Performance LAN Adapter (FC #2980)
>
> I am looking for a layout or a picture of this card showing every detail.
I must have several of these in a box above the garage -- but this is
simply the RS/6000 version of the 3COM 3C527 adapter, so you can
reference http://www.gilanet.com/OhlandL/NIC/3com_527.html for
details, noting that the firmware at 1B and 2B will be different on
the Type 2-1. Yet another "firmware variant" of the 3C527 was used in
the IBM 3172, too (these will have an "orange dot" on the blue plastic
top of the card I/O bracket).
Rick Ekblaw
UZnal
> simply the RS/6000 version of the 3COM 3C527 adapter, so you can
> reference http://www.gilanet.com/OhlandL/NIC/3com_527.html
I did so already, as always the Tool is invaluable when you look for
details.
> for details, noting that the firmware at 1B and 2B will be different on
> the Type 2-1. Yet another "firmware variant" of the 3C527 was used in
> the IBM 3172, too (these will have an "orange dot" on the blue plastic
> top of the card I/O bracket).
I have the 3C527 though not at hand right now, I simply do not know which
(paper) box it is in. It seems as if this type 2-1 card would have a VPD
record, so I must try to dump the VPD data though I am not sure if it is
available on the PS/2 3C527 variant. If I see it right, this is a DMA
capable card.
The interface is pretty simple, commands are mapped to bits and bit groups,
there are some command and status registers starting at the I/O base, there
are mailboxes and so on. It looks a bit like the Spock interface and a bit
like the AHA-154x/164x interface which uses mailboxes.
William R. Walsh
> If I see it right, this is a DMA capable card.
Yes, and it also busmasters. It has an onboard 80188 microcontroller
for that purpose.
I'm not sure it leads to much advantage. My findings on Windows NT
with the card were similar to what Alfred Arnold found on his with
Linux. The IBM LAN/A did much better by the numbers.
William
UZnal
> for that purpose.
>
> I'm not sure it leads to much advantage. My findings on Windows NT
> with the card were similar to what Alfred Arnold found on his with
> Linux. The IBM LAN/A did much better by the numbers.
The Spock206 experiment showed that the driver makes the difference. My
current investigation of this card has another purpose, but look at this
description (from another card) to appreciate that the networking
performance should be tested under different scenarios. With
KeStallExecution(..) and a sufficiently "nice" value (see elink527.sys, NT4
driver, this function inserts delays) one could practically bring every card
to its knees. Besides this, the driver is dated 1996, so I expect it to use
fairly large delays.
"The Inter Packet Gap attribute controls the agressiveness of the adapter on
the network. A smaller number will increase the agressiveness of the adapter
while a larger number will decrease the agressiveness (and increase the
fairness) of the adapter. A smaller number (more agressive) could cause the
adapter to capture the network by forcing other less agressive nodes to
defer. A larger number (less agressive) may cause the adapter to defer more
often than normal. If the statistics for this adapter show a large number of
collisions and deferrals, then try decreasing this number. If the statistics
for other nodes on the network show a large number of collisions and
deferrals, then try increasing this number. Valid values range from 4 to 252
in increments of 4. The default value is 96."
Uli Link
>> Yes, and it also busmasters. It has an onboard 80188 microcontroller
>> for that purpose.
>>
>> I'm not sure it leads to much advantage. My findings on Windows NT
>> with the card were similar to what Alfred Arnold found on his with
>> Linux. The IBM LAN/A did much better by the numbers.
>
> The Spock206 experiment showed that the driver makes the difference. My
> current investigation of this card has another purpose, but look at this
> description (from another card) to appreciate that the networking
> performance should be tested under different scenarios. With
> KeStallExecution(..) and a sufficiently "nice" value (see elink527.sys, NT4
> driver, this function inserts delays) one could practically bring every card
> to its knees. Besides this, the driver is dated 1996, so I expect it to use
> fairly large delays.
Even on old and slow RS/6000 the 2-1 adaptor gives a throughput close to
the theoretical maximum of half-duplex 10MBit/s Ethernet with moderate
CPU usage. So if the busmaster driver (for PC Operating systems) is
written 16MByte RAM ISA busmaster safe the adaptor does a fast
busmastering into a low mem address buffer and then the CPU copies the
data to the destination (a.k.a. double buffering) completely wasting all
benefits of busmastering.
I once have tested the 2-1 with Novell Netware 3.12 and got very good
results.
When the drivers was developed such Ethernet cards were connected to
hubs and not switches. You don't need to fiddle around with fairness on
the Ethernet bus on a switch port. The switch port should generate
"collisions" to slow down the port if the destination cannot handle the
bandwith at the moment.
Those days server adaptors were optimized for combined throughput under
heavy load on a shared media bus.
--
Uli Link
UZnal
It has indeed and is called NMR - network management ROM - on the 3C527 and
VPD on the RS/6000 variant. But read for yourself (the last two pages list
the differences), I got the 3C527, 3C5x9b techrefs and some more docs kindly
provided by
http://www.osdever.net/cottontail/
3C5x9B.PDF list the adapter IDs, the last two are unknown to mcabase:
10BASE2 = 627C
10BASE-T = 627D
Test mode = 61DB (when the adapter is configured for test mode)
TP or coax = 62F6
TP only = 62F7
Hellota...@hotmail.com
the source code based on what they say in section #2.
http://www.usenix.org/publications/library/proceedings/hispeed/full_papers/banerji.a
Ryan
William R. Walsh
> The Spock206 experiment showed that the driver makes the
> difference. My current investigation of this card has another
> purpose
> ... one could practically bring every card to its knees.
If you have any plans to do something on this front, I can test it. In
fact, I can test the daylights out of it...I have a pile of the cards
(including one 2-1) and quite the network environment in which to test
them (many platforms, many types of network, and a link between two
houses).
I was surprised that the card didn't perform any better...the
3C523(TP) and 3C529 are good performers as found although they don't
have the onboard assistance that the 3C527 does. And at one time, 3Com
really stood out in terms of hardware quality. (I suspect that they
still have a pretty good product, but with just about everything
having integrated Ethernet included, their business can't be as good
as it was.)
William
UZnal
> fact, I can test the daylights out of it...I have a pile of the cards
> (including one 2-1) and quite the network environment in which to test
> them (many platforms, many types of network, and a link between two
> houses).
Not now, I am more than busy with other projects.
> I was surprised that the card didn't perform any better...the
> 3C523(TP) and 3C529 are good performers as found although they don't
> have the onboard assistance that the 3C527 does.
The 3C527 is a quite intelligent card, its software model is a textbook
study case. The perfomance will depend on how well you treat the card.
UZnal
The key to the bridge over the troubled 9-K water is this:
"Configuration and RAM/ROM control of PCI extension card residing on MCA
adapter"
http://www.gilanet.com/ohlandl/RS6000/RS6000_9-K.html/
The PCI configuration space (255 bytes or so) of the Am79C971 is accessed
most probably (I hardly hold myself not to say most surely) through the POS
Subextension registers 6 and 7. These accept the byte index of the desired
offset and POS byte 4 provides the window. This should be enabled by the
Adaptec ASIC-9060R bridge and the procedure is described in detail in the
above PDF. An eventual 9-K 8F62 ADF need not be concerned with the PCI
configuration issues and resources.
Note also that the Am79C971 is software compatible with the AMD PCnet family
and LANCE register and descriptor architecture. The NT4 DDK includes a LANCE
sample for the DEC adapters, enough to give some idea but not enough to port
1:1.
As to the 8EF5 9-Q, the AIX configuration method, so I suppose, would take
care of the PCI config issues and set up an interface similar to that of the
original 8EF5 3C527, provided 9-Q really uses the 8EF5 adapter ID (the most
puzzling part of the story).
Both 3C527 and Am79C971 work with two circular receive and transmit rings
(lists) of descriptors. A descriptor holds the length and the physical
address of a buffer in host memory, which is all a DMA busmaster must know
to transfer data.
Large portions of the 8EF5 3C527 AIX driver are provided as a sample in the
AIX 3.2.5. Sorry, I have not noted the exact location, but you can find them
when you start from here
http://stuff.mit.edu/afs/athena/system/rs_aix32
* COMPONENT_NAME: (BOSSAMP) IBM BOS Sample Program
* (C) COPYRIGHT International Business Machines Corp. 1990
* All Rights Reserved
* Licensed Materials - Property of IBM
Ethernet device driver
ciodd.c Common Communications I/O
device interface routines
(top-half driver routines)
cioddi.h Common Comm. I/O header file
ciodds.h Common Comm. I/O header file
cioddhi.h Common Comm. I/O header file
cioddlo.h Common Comm. I/O header file
entds.c device specific routines
entxmit.c data transmission routines
entrecv.c data reception routines
entddi.h device specific header file
entdshi.h device specific header file
entdslo.h device specific header file
ent*.* are the 8EF5 3C527 source files.
/**************************************************************************/
/* POS Register 0 */
/**************************************************************************/
#define POS_REG_0 (0x0) /* POS Register 0 Base offset */
#define PR0_VALUE (0xF5) /* POS Register 0 Adapter ID Code Hex F5 */
/**************************************************************************/
/* POS Register 1 */
/**************************************************************************/
#define POS_REG_1 (0x1) /* POS Register 1 Base offset */
#define PR1_VALUE (0x8E) /* POS Register 1 Adapter ID Code Hex 8E */
Uli Link
> Now I know where I have seen it, I should have a 3C527. But then, the type
> 2-1 card could be assigned to 0041.ADF renamed to 8EF5.ADF, and the drivers,
> provided they do not check the adapter ID, work for the type 2-1 card (if
> not, the adapter ID can be patched in the driver object code). I saw some of
> the 8EF5 driver code, the POS bytes match the 0041.ADF quite exactly.
>
> ADF 0041 3Com EtherLink/MC 32 Ethernet Adapter (3C527) == 8EF5
> 2-1 8EF5 Ethernet High-Performance LAN Adapter [ent] == 0041
> 9-Q 8EF5 10/100 Mbps Ethernet UNI High-Performance Ethernet Adapter == ?
> 9-K 8F62 10/100 Mbps Ethernet MC Adapter (SMP and UNI) [srent, San Remo]
>
> 9-Q and 2-1 share the same adapter ID 8EF5, does that mean they share the
> same AIX driver? Louis claims in the Tool that 9-Q and 9-K are identical,
> was the adapter ID 8F62 perhaps introduced to differentiate between the UNI
> and SMP versions or does it mean more?
>
9-Q and 9-K both have 8f62 and share the same driver package
8-V and 8-U both have 8f95 and share a driver package (also use AMD chipset)
2-1 have 8ef5 is a slightly modified 3c527 and has a different driver
I don't know of any other MC add-on Ethernet adaptors for RS/6000, there
are quite a lot different ID's for different on-board Ethernet.
--
Uli Link
UZnal
8EF5 for 9-Q is listed in "IBM Cluster 1600 Managed by PSSP 3.5: What’s
New":
http://www.redbooks.ibm.com/redbooks/pdfs/sg246617.pdf
Page 207:
MCA-attached hardware
Microchannel bus systems were included in the very first RS/6000 servers and
are not longer included in any pSeries systems. Nevertheless, support by AIX
is included up to AIX 5L Version 5.1. Table D-2 show the available device
drivers.
Restriction: None of the device drivers or machines support the use of the
AIX 5L 64-bit kernel.
Table D-2 Microchannel device drivers
Device number 8EF5
Bus type MCA
Fileset diag rte
AIX Version 4.3.3 0 0
AIX Version 5.1.0 15 0
Feature Code #2964 (9-Q) #2980 (2-1)
Description 10/100 Mbps Ethernet UNI only,
Ethernet High-Performance LAN
> 8-V and 8-U both have 8f95 and share a driver package (also use AMD
chipset)
#2992 (8-U) #2993 (8-V) 10 Mbps Ethernet High-Performance LAN Adapter
> 2-1 have 8ef5 is a slightly modified 3c527 and has a different driver
>
> I don't know of any other MC add-on Ethernet adaptors for RS/6000, there
> are quite a lot different ID's for different on-board Ethernet.
This is the complete Ethernet list I managed to compile and waits to be
included in mcabase:
8EF2 ient_1adapter : sio ent : adapter/sio/ient_1
** Integrated Ethernet for Stillwell type machines
AIX 5L: 7006/7008/7009/7011-250/7012 and SCSI-1 7011-220/230
8EF3 ient_2adapter : sio ent : adapter/sio/ient_2
** Integrated Ethernet for Salmon type machines (see 2-8, 2-9, cf.
AIX 5L)
8F98 ient_6adapter : sio ent : adapter/sio/ient_6
**Integrated Ethernet for Rainbow type machines (10Mbps)
E010 eu adapter : mca eu : adapter/mca/ethernet ??
???
8EF5 (2-1) Ethernet High-Performance LAN Adapter (FC #2980)
8EF3 *(2-8) Integrated Ethernet for Ethernet Riser Cards (Thick/Thin) (FC
#9000 #4221)
8EF3 *(2-9) Integrated Ethernet for Riser Cards (Twisted-Pair) (FC #9001
#4222)
8F95 (8-U) High-Performance (8F95) Ethernet LAN Adapter AUI (10Base5) and
10BaseT (FC# 2992)
8F95 (8-V) High-Performance (8F95) Ethernet LAN Adapter 10Base2 (FC# 2993)
8F62 (9-K) 10/100 Mbps Ethernet MC Adapter (SMP and UNI) (FC #2964 #2994)
8EF5 *(9-Q) 10/100 Mbps Ethernet UNI (FC #2964)
Uli Link
>> 9-Q and 9-K both have 8f62 and share the same driver package
>
> 8EF5 for 9-Q is listed in "IBM Cluster 1600 Managed by PSSP 3.5: What�s
> New":
>
> http://www.redbooks.ibm.com/redbooks/pdfs/sg246617.pdf
Seems to be a documentation error.
<http://www-01.ibm.com/support/docview.wss?rs=1005&context=SSPQKF&dc=DA400&uid=swg27005005&loc=en_US&cs=UTF-8&lang=en&rss=ct1005other>
lists the drivers needed for a given card.
And this is what is needed for my two cards (one 9-Q and one 9-K) too.
No way bringing the 9-Q up without installing devices.mca.9f62
You can remove devices.mca.8ef5 (or it won't be installed on a fresh
install from CD)
--
Uli
UZnal
> > New":
> >
> > http://www.redbooks.ibm.com/redbooks/pdfs/sg246617.pdf
>
>
> Seems to be a documentation error.
That what I start to believe. There is not a single clue in the
devices.mca.8ef5 text about someting else than BNC or DIX.
<http://www-01.ibm.com/support/docview.wss?rs=1005&context=SSPQKF&dc=DA400&u
id=swg27005005&loc=en_US&cs=UTF-8&lang=en&rss=ct1005other>
> lists the drivers needed for a given card.
A quite complete list and errors on a webpage have a higher probability to
be detected and corrected, i.e. this list must be correct.
> And this is what is needed for my two cards (one 9-Q and one 9-K) too.
> No way bringing the 9-Q up without installing devices.mca.9f62
> You can remove devices.mca.8ef5 (or it won't be installed on a fresh
> install from CD)
This the proof.