Same problem here: bad argument when adding tap0.
it seems not due to the "goofy" installation.

Somehow my setup is rather picky about MTU.

I use a Thunderbolt gigabit ethernet which auto-configured its MTU to match the switch one (which happens to be set to the maximum allowed on a 3750: 1998).



when trying to add there was an MTU mismatch which prevented the tap to join

To solve:


and POOF! the succeded

Right now it seems the breakout config NOT to be working (yet) on my current setup, but I will check again later.

Thank you to everybody in pointing to the right direction.

mtg

@mtg

Very well done! How did you track it to an MTU mismatch? While trying to duplicate the problem on my PCs, I used a number of different MTU sizes and was able bridge the tap every time.

I dont have a Mac with a TB interface (soon though). Is it possible that it has somthing to do with the TB interface? While trying solve this for Dan I used my internal en0 and a USB nic and the tap/bridge worked every time.

On a side note:
I dont think you can create a simple DOT1q trunk, or a breakout switch using Mt. Lion. I'll be very surprised anyway. Ive tried every way from Sunday. I have had great success with multiple NICs (one for each VLAN), or using a router configured as a transparent bridge to an ISL trunk. Those methods work like a charm on any system. Ive tested OS X, Windows, Linux, and even FreeBSD.

Best of luck, and please share if you can trunk use a streight up dot1q trunk, or breakout switch.

Jason

_________________
J.Neumann, GNS3 Contributor & Forum Moderator (Minion ID: sin.zx81.1981ad)
Author: Cisco Routers for Small Business (Apress)

mtg

Thanks for the input. I have the Thunderbolt interfaces on my MAC but I don't have the Thunderbolt to Gigabit Ethernet Adapter. Please update on your breakout switch setup so I know it won't be a waste buying the adapter or there is another approach to this.

Thanks Jason as well. Appreciate your help so far.

danviera2002

mtg/Jason,

The MTU fixed my problem, I might not need the Thunderbolt interface for now. I can ping between GNS3 and my physical switch. I will confirm the breakout switch later.

Thanks,
danviera2002

I have verified that it has nothing to do with TB. It's simply an MTU mismatch (as mtg stated) between the Tap interface and the Ethernet interface participating in the bridge. Again, nicely done mtg.

_________________
J.Neumann, GNS3 Contributor & Forum Moderator (Minion ID: sin.zx81.1981ad)
Author: Cisco Routers for Small Business (Apress)

Sorry for my mis-wording.

I meant "I HAPPEN to use a TB-2-Gig adapter, which by itself SELF-ADJUST the MTU to match the one set on the switch".
I regard to this as a behaviour, not as a "feature" ;-).

But since I have a Macbook air, my only two options are either the "no-more-than-MTU1500" usb2ethernet, or
the "jumbo-capable" TB2Gig.
And the latter behaves just as outlined above, and this could be what happened to you.

As Jason outlined (and I agree) this is not related to Thunderbolt itself, but could explain why "it works for somebody, and not for somebody else": it depends on different HW, not necessarily Thunderbolt-based, and how such HW setup manages MTU.
For sure the bridging issue in adding the was due to the MTU mismatch between "en" and "tap".

@Jason,
Since my early networking days I was taught "MTU mismatch is bad", and I luckily discovered that.

Maybe this week I can work on more on my breakout switch setup (which happens to be, due mostly to space and "my wife would kill me if I make a mess of cables and devices" constrains, my only option, see below), and get back to you.


OT: A note on "HW" setup: my current rack is a plain Ikea "LACK" in the "on castors" version.
If you want to install such a piece of furniture, please carefully check airflow of your devices before proceeding:
- usually 3560 and 3750 with Fastethernet have side-to-side airflow, and you should either choose a different "rack" (e.g. the original Lack table), or keep the switches turned on just when you need them (I'd suggest no more than 3-4 hours a day). This is my case.
- usually 3560 and 3750 with Gigabit ethernet have front-to-back airflow and the furniture pictured would fit pretty well.
Further ideas on http://lackrack.org

A little update.

Right now, I managed to let the physical see the virtual switches via CDP.
Hint:
- DISABLE the interface on Network preferences, and manage everything via command line (using ifconfig)

So far:
- Virtual switch correctly tags the cdp frames coming from the virtual router, and the physical breakout switch correctly strip the tag before sending to physical switch
- physical breakout switch correctly tags the cdp frames coming from the physical endpoint switch (i see the frame correctly when capturing the interface), but i seems the frame cannot reach the virtual router.

this on the current dynamips (both RC3 and RC5)
I guess that older versions would work.

Right now, I'll make my last tentative this week:
- use a 3725 w/ 16ESW as virtual breakout switch
- downgrade to older dynamips.

After that, I will turn to linux, and expect (once I solve the thunderbolt gigabit broadcom driver issue: see http://comments.gmane.org/gmane.linux.network/236867 - at worst, I'll use not-apple hardware) to have it working.

Besides, I need the lab to be working for my Cisco CCIE certification, not for Apple certification ;-)

I'll give you some more feedback hopefully this week.

I temporary gave up on OSX.

What I found so far:
1) the bridge correctly hook to the tap (I use the tunnelblick variant).
When tcpdumping the bridge0 interface i can only see packets coming from the tap0, even if both the tap and the en4 (my TB gig) are connected.
This lead me to consider the bridge approach as "unreliable" (at least on OSX 10.8.3), and not related to the gns3.
My next tests will focus on the cleaner approach outlined on http://www.gns3.net/switching-qinq-with-virtual-switch-option/.

2) Just for the sake of test, I changed three gns3 releases: 0.7.4, 0.8.2, 0.8.4-rc2.
None of them seems to be working.
Using the "nio_gen" approach lead (as everybody reading this found) to a core dump of the dynamips instance where the virtual switch is configured

3) using a 3725 with 16ESW show the same core dump.

4) the "nio_gen" approach worked flawlessly on a debian wheezy setup (using the gns3 available on the standard repo: 0.7.4)
I just tested the CDP between phy and logical, and tags are added/removed correctly.
I expect the double tagging to work as well.

5) I tried to contact the author of the article describing the qinq, to ask him/her which OSX/dynamips combination he/she used to test. Waiting for feedback.


Next steps:
- try to obtain the latest dynamips-community code, and compile natively on OSX 10.8.3
- wait for some clever suggestion from either the author or you, my fellow colleagues :-)
- if none of the above working by next week, go for a linux setup, and have the lab (finally) working.

my two cents.

m.

@mtg

Nicely stated, and I think the correct conclusions. Although I use OS X for most of my GNS3 related tasks, I have also decided that Linux provides the least hassel when connecting to physical switches. If you're running OS X Snow Leopard you can connect directly too, but there are still challenges.

As to dynamips on OS X, RC2 comes with two version.
dynamips-0.2.8-RC3-community-OSX.intel64.bin
dynamips-0.2.8-RC5-community-OSX.intel64.bin

RC5 supports the new auto IDLE PC feature.

Best of luck,
-Jason

_________________
J.Neumann, GNS3 Contributor & Forum Moderator (Minion ID: sin.zx81.1981ad)
Author: Cisco Routers for Small Business (Apress)

@ mtg,

Kindly let me know once you have any feedback on the OSX that works. I am still having issues settinh up my lab.

Thanks,
danviera2002