[Atlas-anchors-pilot] Jumbo MTU on service interfaces ?

Romeo Zwart romeo.zwart at ripe.net
Sat Nov 10 14:53:15 CET 2012


Hi Tore,
As mentioned previously, we'll send an configuration recommendation
first half this week.

On 12/11/10 12:10 , Tore Anderson wrote:
> * Nick Hilliard
> 
>> Could you post a desired network configuration for setting up these boxes?
>>  At the moment, it's not clear what exactly you need or what your
>> recommended configuration settings are.
>>
>> I.e.
>>
>> - subnet size / configuration for drac interface
>> - subnet size / configuration for production interface
> 
> Yup. FWIW, my preference for IPv4 on the production interface is /31
> (anything shorter should be justified). On DRAC, my preference would be
> to not require IPv4 at all, or if that's not possible, a /31.

We're still having some internal discussion about IPv4 on the DRAC. We
will anyway not ask for more than a single v4 address on the management
port.

> Unlike IPv4, I have plenty of IPv6 - so assigning up to a /48 is no
> problem. However, I would actually prefer assigning a /127 to the port,
> and routing the assigned prefix using the Anchor as the next hop, in
> order to prevent any neighbour cache exhaustion issues/attacks (cf. RFC
> 6164).
> 
>> - MTU for all
> 
> I don't see the point of discussing jumbo frames. All links leading out
> of my network to the internet has a MTU of 1500 anyway (and I think I'm
> not alone about that), and so have all of the Atlas probes themselves I
> believe. So increasing the MTU on the Anchor will only lead to PMTUD
> having to take place for almost all outbound traffic.

For the anchor and probe functionality I don't think I see a use case
for jumbo frames.

>> - resilient first hop
> 
> Considering that only one production interface is to be used, I don't
> see how this could possibly be achieved?

For the first stage of the pilot, it wouldn't. In a later stage we might
look at using the second production interface for stuff like link
aggregation. This will definitely not be part of the initial roll out,
so not part of the configuration request that we will send out next
week. :)

Cheers,
Romeo





More information about the Atlas-anchors-pilot mailing list