Re: [v6ops] I-D Action: draft-ietf-v6ops-ipv6-ehs-packet-drops-01.txt

[Brian E Carpenter <[e-mail]>]

On 15-Oct-20 05:39, Fernando Gont wrote:
...
>> I am not sure if I agree with the last paragraph of 6.1 which opines on
>> the low adoption of rfc6437-style flow label hash entropy. 
> 
> We don't argue low-entropy, but issues with the Flow  Label 
> implementation. See e.g.: 
> https://blog.apnic.net/2018/01/11/ipv6-flow-label-misuse-hashing/
> 
> 
> 
>> I would
>> expect that the current low adoption in ecmp implementations is a
>> chicken/egg problem.  If you have a product that needs to hash and you are
>> not confident there is enough adoption in host stacks, you are still forced
>> to hunt for entropy from the upper-protocol protocols.  Otherwise all the
>> legacy host traffic with the flow label set to 0 risks getting hashed
>> entirely to one side or face reordering.
> 
> We're just stating that there's marginal usage of the Flow Label, which 
> is what the work by Cunha et al seems to indicate. As to the possible 
> consequences, part might be LBs lagging behind host implementations. But 
> issues in the Flow Label implementation might have also discouraged its use.
> 
> That said, the important info that Section 6.1 is meant to convey is 
> that there still seems to be marginal use of the FL. And also note that 
> issues have been found in FL implementations. -- i.e., we're not trying 
> to infer why there's marginal use of the FL.
> 
> If you think this warrants some tweaks in the text, please do let us know.

I just re-read that section and sadly I think it is still correct.

   Brian (co-author of several flow label RFCs)