Skip to contentSkip to conversation

Linux 2.2.18pre21

DISCUSSION willy tarreau 6 participants 10 messages Nov 10, 2000 – Nov 11, 2000

lore
Share & tools
lore
#0 lore raw
Quoted 2 lines

Anything which isnt a strict bug fix or previously agreed is now 2.2.19 material.

Alan, do you consider it as a bugfix if I tell you that
we can't get anymore oops with the new bonding code, even in SMP ?

I've had reports of it working very well, and faster, for a long time now and the link detection seems completely OK now. I'd like to specially thank Constantine Gavrilov for all the tests he has done and the time he spent in enhancing the documentation.

Please find the patch against 2.2.18pre20 in
attachment, in case you agree.

Regards,
Willy

___________________________________________________________
Do You Yahoo!? -- Pour dialoguer en direct avec vos amis, Yahoo! Messenger : http://fr.messenger.yahoo.com

#1 lore raw

On Fri, Nov 10, 2000 at 10:28:46AM +0100, willy tarreau wrote:

From the patch source:

+CONFIG_BONDING

+  Say 'Y' or 'M' if you wish to be able to 'bond' multiple Ethernet
+  Channels together. This is called 'Etherchannel' by Cisco,
+  'Trunking' by Sun, and 'Bonding' in Linux.

I think "bonding" is term used in one particular type of ISDN multilink calls.

Cisco Trademark is EtherChannel -- there the capitalization is important. We could call it ETHERNETCHANNEL (and even "Etherchannel" or "ETHERCHANNEL") get away with it clean.

...

Quoted 2 lines

Regards,
Willy

/Matti Aarnio
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

#2 lore raw
In reply to willy tarreau
Quoted 21 lines · Matti Aarnio

On Fri, Nov 10, 2000 at 10:28:46AM +0100, willy tarreau wrote:

From the patch source:

(quoted diff) +4 -0
+CONFIG_BONDING
+  Say 'Y' or 'M' if you wish to be able to 'bond' multiple Ethernet
+  Channels together. This is called 'Etherchannel' by Cisco,
+  'Trunking' by Sun, and 'Bonding' in Linux.
 
        I think "bonding" is term used in one particular type
        of ISDN multilink calls.
 
        Cisco Trademark is  EtherChannel -- there the capitalization
        is important.  We could call it ETHERNETCHANNEL (and even
        "Etherchannel" or "ETHERCHANNEL") get away with it clean.
 
...
Quoted 2 lines

Regards,
Willy

Quoted 1 line

/Matti Aarnio

ISDN uses "channel bonding", not bonding. As for "Etherchannel", let us change it to "EtherChannel" is this is how it is called.

Signature · 12 lines
----------------------------------------
Constantine Gavrilov
Unix System Administrator and Programmer
Xpert Integrated Systems
1 Shenkar St, Herzliya 46725, Israel
Phone: (972-8)-952-2361
Fax:   (972-9)-952-2366
----------------------------------------
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/
#3 lore raw
In reply to willy tarreau
Quoted 11 lines · Constantine Gavrilov
Quoted 8 lines

Cisco Trademark is EtherChannel -- there the capitalization is important. We could call it ETHERNETCHANNEL (and even "Etherchannel" or "ETHERCHANNEL") get away with it clean.
...

Quoted 2 lines

Regards,
Willy

/Matti Aarnio

ISDN uses "channel bonding", not bonding. As for "Etherchannel", let us change it to "EtherChannel" is this is how it is called.

Anything but "EtherChannel" -- trademark people are sometimes unpleasant when they consider something being infringed. (And nowhere as much as in USA..)

We dont' have cisco approval of using their trademark in Linux kernel feature name, or do we ?

Quoted 4 lines
Signature · 3 lines
Constantine Gavrilov
Xpert Integrated Systems
1 Shenkar St, Herzliya 46725, Israel

/Matti Aarnio
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

#4 lore raw
In reply to willy tarreau
Quoted 15 lines · willy tarreau
Quoted 3 lines

Anything which isnt a strict bug fix or previously agreed is now 2.2.19 material.

Alan, do you consider it as a bugfix if I tell you that
we can't get anymore oops with the new bonding code, even in SMP ?

I've had reports of it working very well, and faster, for a long time now and the link detection seems completely OK now. I'd like to specially thank Constantine Gavrilov for all the tests he has done and the time he spent in enhancing the documentation.

Quoted 5 lines

Please find the patch against 2.2.18pre20 in attachment, in case you agree.

Regards,
Willy

I do think the new code is much better. Anyone who is considering using trunking for their projects should use the new code since it has link monitoring and active backup mode. However, I am a maniac when it comes to kernel testing. If I say it really works, it usually works. This code has been tested much and bugs were fixed. However, it has not been tested enough that I may bet by head on saying there are no known issues. This is because I did not have access to all hardware that was needed to complete the tests in time.

The code is not worse than the old code. However, a clear note must be made that this is experimental code and probably has small issues. So if it is included, it must me marked as experimental.

----------------------------------------
Constantine Gavrilov
Unix System Administrator and Programmer
Xpert Integrated Systems
1 Shenkar St, Herzliya 46725, Israel

Phone: (972-8)-952-2361
Fax:   (972-9)-952-2366

----------------------------------------
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

#5 lore raw
In reply to willy tarreau
Quoted 14 lines · Matti Aarnio
Quoted 11 lines · Constantine Gavrilov
Quoted 8 lines

Cisco Trademark is EtherChannel -- there the capitalization is important. We could call it ETHERNETCHANNEL (and even "Etherchannel" or "ETHERCHANNEL") get away with it clean.
...

Quoted 2 lines

Regards,
Willy

/Matti Aarnio

ISDN uses "channel bonding", not bonding. As for "Etherchannel", let us change it to "EtherChannel" is this is how it is called.

Quoted 7 lines

Anything but "EtherChannel" -- trademark people are sometimes unpleasant when they consider something being infringed. (And nowhere as much as in USA..)

We dont' have cisco approval of using their trademark in Linux kernel feature name, or do we ?

Gee, we do not call it EtherChannel, we say CISCO calls it EtherChannel. Where is the infringment here? Are people that paranoid or it is just me who is not getting it?

Signature · 12 lines
----------------------------------------
Constantine Gavrilov
Unix System Administrator and Programmer
Xpert Integrated Systems
1 Shenkar St, Herzliya 46725, Israel
Phone: (972-8)-952-2361
Fax:   (972-9)-952-2366
----------------------------------------
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/
#6 lore raw
In reply to willy tarreau
Quoted 3 lines · Constantine Gavrilov

Gee, we do not call it EtherChannel, we say CISCO calls it EtherChannel. Where is the infringment here? Are people that paranoid or it is just me who is not getting it?

You missed my original point.

I don't like to call it BONDING.

"Bonding" is something where two (or more) channels carry data in between two participating systems. Like Multilink-PPP, and ISDN Channel Bonding. Often indeed data goes out somehow inter- leaved on the physical links. (Like ISDN Channel Bonding supplies a transparent 128 kbps link instead of two 64 kbps links to the upper layers.)

EtherChannel does select the link (out of the group) by forming XOR of source and destination MAC addresses (their lowest bytes), and then doing MODULO number-of-links on the result.

So between systems A and B the flow goes via link 0, in between A and C it goes via link 1. Add there client system D, and it may end up into either of the links.

|-----------| |------|
| A |-----link-0-----| SW |---[B]
| |-----link-1-----| with |---[C]
|-----------| |EthChn|---[D]
|------|

This gives improved throughput on congested links in between two switches, or major server and core switches, while preserving data order over the links.

Blind bonding-type "throw packets on links 0 and 1" MAY end up sending ethernet frames out of sequence, which for a few LAN based protocols is a great source of upset.

Beowulf systems have "bonding" in use for parallel Ethernet links in between two machines, however THAT is not EtherChannel compatible thing!

Quoted 2 lines
Signature · 1 line
Constantine Gavrilov

/Matti Aarnio
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

#7 lore raw
In reply to willy tarreau
Quoted 4 lines · Matti Aarnio

Beowulf systems have "bonding" in use for parallel Ethernet links in between two machines, however THAT is not EtherChannel compatible thing!

Maybe we should adopt's sun naming then, and call it 'Trunking'.

This is the same driver that Beowulf uses, and it is Etherchannel compatible.

The only part of Etherchannel we don't support is the XOR channel selection (yuck!) and the automatic configuration of the links (it's a MII thing, that's undocumented.)

Leave it as Ethernet Bonding.

Signature · 8 lines
------------------------+--------------------------------------------------
Thomas Davis		| PDSF Project Leader
[email protected]		| 
(510) 486-4524		| "Only a petabyte of data this year?"
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/
#8 lore raw

Intel's ANS Driver -vs- Bonding [was Re: Linux 2.2.18pre21]

In reply to willy tarreau

I think it is great that there is continued valuable developement on the bonding driver. Have you guys taken a look at the source code for Intel's new ANS driver? For any Intel network card, it will do 8-way Fast EtherChannel. Supposedly, it also supports failover (though even "bonding" driver docs used to say that was impossible because the linux networking subsystem didn't handle card failures gracefully enough).

You can check it out at:

http://support.intel.com/support/network/adapter/pro100/100Linux.htm

Maybe some of the code will help in the "bonding" driver development. I much prefer bonding over Intel's ANS, because bonding is GPL and works with any net card. Etc. etc.

-Dan

Quoted 17 lines · willy tarreau
Quoted 2 lines

Anything which isnt a strict bug fix or previously agreed is now 2.2.19 material.

Alan, do you consider it as a bugfix if I tell you that
we can't get anymore oops with the new bonding code, even in SMP ?

I've had reports of it working very well, and faster, for a long time now and the link detection seems completely OK now. I'd like to specially thank Constantine Gavrilov for all the tests he has done and the time he spent in enhancing the documentation.

Please find the patch against 2.2.18pre20 in
attachment, in case you agree.

Quoted 7 lines

Regards,
Willy

___________________________________________________________
Do You Yahoo!? -- Pour dialoguer en direct avec vos amis, Yahoo! Messenger : http://fr.messenger.yahoo.com

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/

#9 lore raw

Re: Intel's ANS Driver -vs- Bonding [was Re: Linux 2.2.18pre21]

Quoted 8 lines · Dan Browning

I think it is great that there is continued valuable developement on the bonding driver. Have you guys taken a look at the source code for Intel's new ANS driver? For any Intel network card, it will do 8-way Fast EtherChannel. Supposedly, it also supports failover (though even "bonding" driver docs used to say that was impossible because the linux networking subsystem didn't handle card failures gracefully enough).

Quoted 7 lines

You can check it out at:

http://support.intel.com/support/network/adapter/pro100/100Linux.htm

Maybe some of the code will help in the "bonding" driver development. I much prefer bonding over Intel's ANS, because bonding is GPL and works with any net card. Etc. etc.

Alas, the license has the BSD documentation requirement, and thus is incompatible with the GPL...

Jeff

Signature · 7 lines
Jeff Garzik             |
Building 1024           | Would you like a Twinkie?
MandrakeSoft            |
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
Please read the FAQ at http://www.tux.org/lkml/