Posted on February 18, 2016 12:12 am
 |  Asked by Fredrik Pettai
 |  19660 views
0
0
Print Friendly, PDF & Email

Hi,

How is the Jumbo frame MTU calculated in Arista vs. Cisco vs. Juniper?

Max MTU in Arista -> 9214 vs. Cisco -> 9216 vs. Juniper -> 9192

They should all be the same frame size on the wire, right? (it’s a bit confusing that they are displayed differently in their respective OS…)

Ex.

EOS> show interfaces ethernet 1 | grep MTU

IP MTU 9214 bytes , BW 10000000 kbit

EOS> show interfaces ethernet 1 l2 mtu

Port        MTU Config MTU Status

———-  ———- ———-

Et1                          9234

 

JUNOS> show interfaces xe-0/1/8 | grep MTU

Link-level type: Ethernet, MTU: 9192, MRU: 0, LAN-PHY mode, Speed: 10Gbps,

Protocol inet, MTU: 9178

Protocol multiservice, MTU: Unlimited

 

0
Answered on February 18, 2016 10:00 am

Hello

I can be wrong :)

Point 1 : there is no standard regarding what is the size of a Jumbo frame (Layer 2). Arista, Cisco, Juniper …. are saying it is around 9k

Point 2 : We can define the MTU from a Layer 2 perspective or a Layer 3 perspective

From you example you have 9214 and 9234

9214 is the IP MTU and 9234 is the Layer 2 MTU. It make sense because the IP header is 20 bytes.

0
Posted by Chris Kane
Answered on February 18, 2016 7:27 pm

There doesn’t appear to be a defined standard to use when Jumbo Frames are discussed.

Here’s a link to an IETF submission regarding the use of Jumbo Frames at IXPs.

https://tools.ietf.org/html/draft-mlevy-ixp-jumboframes-00

You can see in section 1.2 that ”Jumbo Frames are considered to be Ethernet frames that can carry an IP payload greater than 1,500 bytes.”

Here’s a link to a discussion back in 2011 regarding what MTU size to use at IXPs.

https://www.ietf.org/proceedings/82/slides/grow-2.pdf

In my experience (disclaimer; YMMV), the only time the exact MTU value matters and must match is when running OSPF. As John Moy points out in his book OSPF Anatomy of an Internet Routing Protocol, they found that during the development of OSPF the various flavors of transport (ex. TokenRing, FDDI) was causing them problems during the Database Description exchange. So they designed OSPF to detect and avoid links having MTU mismatches.

I hope that helps,

-chris

 

 

0
Posted by John
Answered on February 26, 2017 7:30 pm

tl;dr : Arista==Cisco==(Juniper – all overhead)

I think the confusion comes in the vendor’s choice in assumptions for displaying Maximum Transmission Unit (MTU). For simplicity, I will stick to EtherType the IP protocol and not include other encapsulations (or encapsulation combinations).

Media MTU = Encapsulation Overhead + Protocol MTU

Transmission Unit (up to the MTU):

  • Non-Jumbo
    • Untagged: 64-1518 = 6 (Source MAC) + 6 (Destination MAC) + 2 (EtherType) + 46-1500 (IP Payload) + 4 (FCS)
    • 802.1Q Tagged: 68-1522 = 6 (Source MAC) + 6 (Destination MAC) + 2 (EtherType) + 4 (802.1Q Tag) + 46-1500 (IP Payload) + 4 (FCS)

Vendor Configuration:

  • Arista Networks
    • Configuration of Interface MTU is the Maximum IP Payload Size
  • Cisco Systems
    • Configuration of Interface MTU is the Maximum IP Payload Size
  • Juniper Networks (may vary from hardware/OS/versions):
    • Configuration of the Interface MTU is the Maximum Media Payload Size
      • IP Payload + All Overheads
    • Some platforms you may be able to configure the Protocol MTU on the sub-interface
    • IP protocol MTU can be shown by retrieving the operational output of the IP subinterface
    • Also note: SRX requires additional overhead to be taken into account for Redundant Ethernet Interfaces
  • Max configurable MTU will vary from vendor to vendor as well as different software versions and/or platforms and/or interfaces even within the same hardware vendor.

Obviously, this doesn’t take into account to host OS’s varying calculations of MTU… but typically it is the Maximum IP Payload Size

Helpful References:

Post your Answer

You must be logged in to post an answer.