On Wed, Feb 17, 2010 at 5:23 PM, Salman  Abdul Baset
<[email protected]> wrote:
> On Wed, 17 Feb 2010, Eric Rescorla wrote:
>
>> On Wed, Feb 17, 2010 at 4:38 PM, Salman  Abdul Baset
>> <[email protected]> wrote:
>>
>>> Unless I misread your earlier post on this list, there may be a
>>> possibility
>>> of a collision attack. Although, as you pointed, it is a preimage
>>> resistance
>>> issue.
>>
>> Again, huh?
>>
>> The collision resistance of a b bit hash function is approximately
>> 2^{b/2}.
>> The collision resistance of the first (or last, or any) b' bits of a b bit
>> hash function is approximately 2^{b'/2}.
>>
>>
>>
>>> Rather than arguing about hash function properties, I think the most
>>> future-proof solution is:
>>>
>>> -fix a hash function for the mandatory-to-implement DHT, which is already
>>> there in the draft.
>>> -allow a large size hash function and corresponding length node-IDs to be
>>> used per overlay, by specifying them in the configuration file.
>>
>> I don't understand why this is superior. Again, what's wrong with just
>> having
>> fixed size node-IDs which are created by truncating a hash?
>
> Flexibility and future-proofing.
>
> If we want to keep the 128-bit node-IDs, then we need to clarify in the
> draft that they can also be obtained by trucating a >128-bit hash function
> without worrying about collisions. For example, this will be helpful for
> someone who may notice the node-ID length difference between the original
> Chord paper and RELOAD Chord. We also need to clearly state that such a
> truncation is applicable for any DHT (Kademlia, Pastry, CAN) and not just
> Chord.

I have no problem with adding this clarification.

-Ekr
_______________________________________________
P2PSIP mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/p2psip

Reply via email to