Jumbo frame question/confirmation

mark_s0

Hi, I'm currently reading into jumbo frames and I was hoping someone could confirm what I think I've understood from it all so far and possibly answer some of my questions.

- Jumbo frames are a layer 2 technology.

- They are frames larger than 1500bytes that can only be used on gigabit interfaces and have to be supported on both ends.

- I've read that Jumbo frames sent to 10/100 interfaces are dropped but then I've also read that Path MTU Discovery (Layer 3) allows the two devices to negotiate the MTU through ICMP messages. I'm guessing that the Path MTU Discovery stage would happen over a WAN link or between routers where as the dropped frames scenario would be on a LAN? I also read that the MTU is negotiated during the TCP handshake (MSS+40bytes headers) so if the MTU is already set before the data transmission starts, surely you wouldn't need PMTU? Maybe I've misunderstood that part.

- Jumbo frames sent over a WAN link can be sent as a whole packet if the routers and WAN link support that MTU size.

- If the routers and WAN link's don't support it and the DF bit is set in the packet, then the packet will be dropped. What scenario would it be set or not set?

- If the WAN link's do support it and the DF bit isn't set in the packet, then the packet will be fragmented. If the packet is fragmented, how is the packet then reliably reassembled at the receiving end?
I understand that TCP uses acknowledgements, is this used in the reassembly phase?

I'm only really at CCENT level knowledge so this is probably beyond my current level and i've possibly confused myself/fried my brain but i'd appreciate any feedback or further resources on the topic, if only for pure interest at the moment

Thanks,

Mark

Forsaken_GA

mark_s0 wrote: »

Hi, I'm currently reading into jumbo frames and I was hoping someone could confirm what I think I've understood from it all so far and possibly answer some of my questions.

- Jumbo frames are a layer 2 technology.

Well, yes, seeing as how frames are a layer 2 encapsulation.

- I've read that Jumbo frames sent to 10/100 interfaces are dropped but then I've also read that Path MTU Discovery (Layer 3) allows the two devices to negotiate the MTU through ICMP messages. I'm guessing that the Path MTU Discovery stage would happen over a WAN link or between routers where as the dropped frames scenario would be on a LAN? I also read that the MTU is negotiated during the TCP handshake (MSS+40bytes headers) so if the MTU is already set before the data transmission starts, surely you wouldn't need PMTU? Maybe I've misunderstood that part.

Ok, there's a few bad assumptions here. Maximum Segment Size is not always equal to MTU. If the remote side wants an MSS of 64k, and you agree to it, that's the size your TCP segments will be. However, the MTU won't support encapsulating that much data at once, so the outgoing packets get fragmented.

However, let's say for arguments sake that MSS does equal MTU. Well, the remote is probably going to use it's MTU. The problem is that it knows nothing of the MTU on the path between the requestor and itself. So let's say they negotiate an MSS of 1500, standard ethernet frame size. Except, oops, one of the links in between them has an MTU of 1496. When the 1500 frames get sent to that link, they'll be fragmented and then sent on, unless the DF bit is set, in which case, they'll just be dropped.

This is why Path MTU Discovery is useful. It figures out what is the maximum supported MTU end to end, ie, which link has the smallest MTU. Then both sides can use that for their MSS to ensure nothing gets fragmented in transit.

- If the routers and WAN link's don't support it and the DF bit is set in the packet, then the packet will be dropped. What scenario would it be set or not set?

- If the WAN link's do support it and the DF bit isn't set in the packet, then the packet will be fragmented. If the packet is fragmented, how is the packet then reliably reassembled at the receiving end?
I understand that TCP uses acknowledgements, is this used in the reassembly phase?

For these two answers, I'd recommend consulting tcpipguide.com (mainly because I don't feel like doing that much typing )