Routing

Hi all,

I am in process of allowing Jumbo frames on the uplink (mtu = 9000) towards the provider. The provider has already configurged the new mtu limit towards my router. I have few questions related to this:

1. During the configuration, will there be any BGP flap etc? Should I bleedoff the traffic OR mtu config will not have any production impact?
2. In order to get full utilization of Jumbo frames, my understanding is that I have to confugure the new mtu limit between all the following links:-

Edge Router <-> aggregation switch <-> ToR <-> Server 

Is this correct?

Thanks

Hello,

---

@vakas10 wrote:

Hi all,

 

I am in process of allowing Jumbo frames on the uplink (mtu = 9000) towards the provider. The provider has already configurged the new mtu limit towards my router. I have few questions related to this:

1. During the configuration, will there be any BGP flap etc? 

For single-hop eBGP with link IP peering - yes. JUNOS flaps the interface when link MTU is changed.

For iBGP - no, if You have a redundant link and You change one link MTU at a time

---

@vakas10 wrote:


Should I bleedoff the traffic OR mtu config will not have any production impact?

 

 

There most likely WILL be a production impact or even multiple. You need to take some precautions to mitigate them.

To give You a few examples:

- OSPF will flap adjacencies when MTU is changed either way

- ISIS most likely not when MTU is inflated although I haven't tested it in recent releases

- BGP : see above

---

@vakas10 wrote:

1.  
2. In order to get full utilization of Jumbo frames, my understanding is that I have to confugure the new mtu limit between all the following links:-

Edge Router <-> aggregation switch <-> ToR <-> Server 

Is this correct?

 

Yes. But only for that stretch - internet traffic won't be able to use Your new MTU.

HTH

Thx

Alex

Hi vakas10,

It's great to be cautious with MTU change on interfaces, it does cause IFL/IFA flap (logical interface and primary L3 interface address) for reprogramming.  While this might not impact BGP session that's not using the interface address but you can expect other protocols like OSPF/ISIS or even LDP (that I'm aware) to flap for sure, and even BGP if its on the interface address.

Just in case the uplink is a LAG, ensure we set MTU on the LAG and not the members:

https://kb.juniper.net/InfoCenter/index?page=content&id=KB22639&cat=JUNOS_PLATFORM&actp=LIST

With regards to utilizing jumbo frames, yes its best to enable it end-to-end if the servers can support it.

Hope this helps.

Regards,
-r.

--------------------------------------------------

If this solves your problem, please mark this post as "Accepted Solution."
Kudos are always appreciated :).