Re: [TLS] New directions in certificate status

[Rob Stradling <[e-mail]>]

On 09/10/14 16:07, Watson Ladd wrote:
> On Wed, Oct 8, 2014 at 6:07 PM, Phillip Hallam-Baker
> <[e-mail]> wrote:
>> Please note:
>>
>> http://datatracker.ietf.org/doc/draft-hallambaker-compressedcrlset/
>> Also note the pending IPR disclosure.
>>
>> In brief Rob Stradling and myself have come up with a radically new
>> approach to certificate status that is vastly more efficient than any
>> previous proposal that provides finer grain certificate status than
>> the certificate validity interval.
>
> Good idea. Unfortunately the draft is incomprehensible: key details
> are missing. If understand this correctly, the idea is to use a
> crit-bit tree to encode the certificate serial numbers, by removing
> common prefixes. There are a lot of finicky details to specify, but
> this should help.
>
> What would be nice is some hard numbers on how much it helps.

Hi Watson.  I won't comment on the methods right now, but here are some 
hard numbers from my experimental code...

Using a snapshot of Google's Pilot/Aviator CT logs from mid-June, I 
counted 2,616,669 unexpired-and-publicly-trusted certs useable for TLS 
Server Authentication.  I grabbed all of the applicable CRLs that were 
available online and found that 286,875 of these certificates were revoked.

Size of the resulting "Compressed CRLSet": ~182K

That's ~6 bits per unexpired-and-revoked-and-publicly-trusted TLS Server 
Certificate.
Or, to make it sound more impressive, ~0.6 bits per 
unexpired-and-publicly-trusted TLS Server certificate.

<snip>

-- 
Rob Stradling
Senior Research & Development Scientist
COMODO - Creating Trust Online