Standard MTU for SJFs and support for MEF 3.0 AI payloads
Hi Friends! Everyone loves debating MTU sizes, right?! With the upcoming new MEF3.0 (AI) standard potentially requiring support for higher than 9100 byte payloads, I'm trying to decide on a good new "standard" to configure on internal backbone/IGP links. We're effectively talking about "Super Jumbo Frames", or SJFs, looking to support at least 9400 payloads (to be future proof). With previous hardware generations limiting most of us to 9216, a lot of ISPs chose 9192 as the base Ethernet MTU (= 9178 IP), allowing for simple underlays (e.g. vlan-tunneling IGP connections through L2 switches) if necessary. While it's of course only a requirement to keep a consistent MTU across your own network, there are benefits to using industry standardized value(s), as some of the current generation hardware limits the number of different MTU values you’re able to configure in parallel on the same NP. Looking at my own network, I have a hard limit right now at 9646 bytes (NCS 5700, Jericho2). Potential new MTUs I'm considering: - 9622 (supporting 24 bytes underlay, like we had for 9192) - 9530 (supporting 116 bytes underlay, same as MEF3.0 [9100] has from 9216) With us looking at deploying SRv6, looking at an IPv6 SRH with 4 SIDs: Ethernet (14) [+ VLAN (4)] + IPv6 (40) + SRH (72) = 130 bytes total By chance this is matching the total frame size with a customer payload of 9400. Is 9530 a good value? What would you configure? (Am I simply overthinking this?) BR, // CF
On Mon, 9 Mar 2026, Carl Fredrik Lagerfeldt via NANOG wrote:
Is 9530 a good value? What would you configure? (Am I simply overthinking this?)
I did a very brief overview of this a few months back, and found several references to 9600. https://business.comcast.com/support/article/ethernet/wavelength-service being one of them. "Transport interfaces can support frame sizes of up to 9600 MTU. " Not sure what "frame size" is in this context exactly, if's calculated with 14 bytes of ethernet header or something else. https://www2.arelion.com/ethernet-vs-waves-flyer , 1299 also say 9600 on DWDM (but it seems you're now saying you could support more?) I don't think we generally can agree on a frame size in this industry, 4470 and 9180 were the IP MTUs of classical old transport, but nothing say we should stay there. So 9530 sounds as good as any. Just define it well so everybody knows what's up. I've tried asking several providers what their number means, if it's L2 or L3 MTU they're mentioning, and generally the provisioning/installation people don't know the difference so I've had to empirically test what it was between those two variants. -- Mikael Abrahamsson email: swmike@swm.pp.se
On Mon, Mar 09, 2026 at 07:25:11PM +0000, Carl Fredrik Lagerfeldt via NANOG wrote:
Hi Friends! -snip- Is 9530 a good value? What would you configure? (Am I simply overthinking this?)
Given the long tail, and life of equipment in carrier Ethernet networks in particular, I'd probably still be configuring 9192-9216ish in many environments. Additional tag space could be nice*, but few have the luxury of greenfield builds where all the equipment is new. Folks are typically working with a least common denominator around 9192 to 9216, just in case a frame has to traverse something unexpected. Most customers are well conditioned today to ask for 9000, but not neccesarily more. * It's probably also worth noting that as long as equipment vendors only support N nested tags/labels, additional MTU might not be a huge help by itself. --msa
On 09/03/2026 22:48, Majdi S. Abbas via NANOG wrote:
* It's probably also worth noting that as long as equipment vendors only support N nested tags/labels, additional MTU might not be a huge help by itself.
I tend to agree. I don't think there is any practical utility in raising the jumbo frame MTU by MEF's new standards given not only the current state of equipment, but also, the current state of MTU operations in the Internet's networks. Mark.
participants (4)
-
Carl Fredrik Lagerfeldt -
Majdi S. Abbas -
Mark Tinka -
Mikael Abrahamsson