Hi there

I have a PCC balancer running QinQ (vlan inside vlan, if I understand it correctly.
Should I change the default MTU for the Ethernet ?
Setup is as follows:
Router 1:
ether1 - default MTU
vlan1 - interface=eth1
vlan101 - interface vlan1
IP address assigned to vlan101
I have many interfaces configured this way. Also, do I need to use service tag on the "main" vlan?

Any help would be appreciated.
Thanks
W

Common advice is to increase the MTU from 1500 to 1504 bytes. But I'm not sure if on ROS it's the L2MTU you'd be looking at instead. I'm gonna do some fragmentation tests tomorrow and see what I can find.

The service tag is usually used on the outside and sets the packets' TPID to 0x88a8, signalling that this vlan is a s-vlan (think of "s" as in "service provider") containing c-vlans (c = customer) according to 802.1ad. The normal vlan TPID (802.1q) is 0x8100 and remains the same for c-vlans.

Wether you need to set the service tag depends on your setup. If you use q-in-q with devices from other manufacturers chances are they can only deal correctly with your s-vlan if the TPID is correctly set to 0x88a8. But on a lot of devices the inner and outer TPIDs can be configured.

We use it this way because
a) there's no reason for us not to comply to the standard and
b) our carrier expects correctly marked 802.1ad packets, strips the s-vlan tag and delivers the c-vlan packets to our customers.

P.S.:
802.1ad was developed out of 802.1QinQ, often confused with 802.1q. In 802.1QinQ 0x8100 was used. This was changed in 802.1ad. So maybe being able to still use 0x8100 as outer TPID may have to do with backwards compatibility. See http://en.wikipedia.org/wiki/IEEE_802.1ad

Q-in-Q and S-tags are slightly different animals - As AlArenal pointed out, S-tags are used more commonly by carriers - q-in-q without an s-tag is liekly what you're looking for.

You will incur another 4 bytes of overhead in the L2 MTU since this is additional overhead in the Ethernet frame header.

Guys, thanks so much for this, but I need more 'lamens' terms here. My current setup is like this:

rb1100 - 01
eth1
vlan1
vlan101
vlan102
vlan103
vlan104


Other side router, which I also have:

rb1100 - 02

eth12
vlan1
vlan101
vlan102
vlan103
vlan104

I have to also bridge at router 2, since the goal is to bring the second rb100's ether port to the main router in virtual format, for the use of PCC balancing over many lines.

So I have:

br1
eth1
vlan101
br2
eth2
vlan102
br3
eth3
vlan103

So in my simple mind, I would think I need service tags in the main vlan trunk, where all the trunks underneath it also sits ?

If anybody has a better idea for bridging those ethernet in as layer2 but virtual, please help. This was my first idea, but I am not sure if its the best.

Thanks!

First of all read this:
http://wiki.mikrotik.com/wiki/Manual:Ma ... uterBoards

Each interface on RouterBOARD in RouterOS have 3 values
1) MTU
2) L2 MTU
3) Max L2 MTU (read-only)

Golden rule:
DO NOT TOUCH MTU OF PHYSICAL INTERFACES - there are really, really rare situation when you really have to do that.

in your situation:
if "L2 MTU minus MTU" is enough to add additional VLAN header - you do not have to change anything
If "L2 MTU minus MTU" is not sufficient for additional header
a) you can increase L2 MTU, but not more than Max L2 MTU
b) you can reduce MTU, but then you have to make sure that you have change-mss in place and opposite end of wire also have same adjustment.

Thank macgaiver

Thats very useful to know. It would seem that Mikrotik changes these L2MTU accordingly to accomodate for the extra header.

My other question still remains; how does the Service tag link into this ? I have Mikrotik on both sides of the vlan trunks.

"use-service-tag" is like a language switch, it enables RouterOS devices to speak in http://en.wikipedia.org/wiki/IEEE_802.1ad

It's up to you to use it or not. Technically it doesn't really make difference in your setup if you use double tagged dot1q or s-vlans.

I'm not sure about how regular layer 2 switches handle dot1ad packets but I assume that they should handle the outer tag of a double tagged dot1q just fine - as long as you don't use regular dot1q vlans with the same vids on the same device.

Many thanks guys !

Golden rule:
DO NOT TOUCH MTU OF PHYSICAL INTERFACES - there are really, really rare situation when you really have to do that.

What if we want the benefit of large mtu ?
I have 3 PC routers connected to a cisco 2960 . I was thinking to increase mtu of switch and then increase the mtu (or maybe l2mtu or both ?!) of interfaces of routers to 9000 . router interfaces are RB44Ge . so you suggest not to do ?

1) Whole path from source to destination need to support your MTU, (both client PC and server) (whatever it is 1500 or 9000)

2) if at least one device doesn't support this value you will need change-mss ruls for TCP and will have to live with fragmentation for other traffic hoping that fragments are not dropped somewhere in the firewall.

So sure, you can have your whole local network have MTU to 9000 and then have change-mss rules on the gateway, but currently i do not see the benefits that is worth the effort

I just want the benefit of mtu 9000 between my 3 routers . I already use tcp mss 1420 because most users are PPPoE clients over mtu 1500 . Upstream providers mtu is also 1500 .
According to switch mtu=9000 , which parameters should change and to what ?
mtu ? l2mtu ? value ?