At Wed, 22 Oct 2008 09:38:07 +1000, Terry Manderson wrote:
> 
> On 20/10/2008, at 8:22 PM, Andy Newton wrote:
> 
> > Hmmm... This couldn't be too hard to test, now could it.  Using
> > curl or Apache libraries should be easy.  If somebody could throw
> > out what they consider to be a good size for the number of files
> > in the repository and file size ranges, I'll cook something up.
> 
> APNIC _had_ a test repository in place. I don't know if it is still
> there.  I'm not aware of any other organisation with a populated
> repository of test objects.

APNIC did indeed have a mock-up of a full-scale RPKI repository at one
point, based on available data from IANA and all five RIRs (ie, this
was an attempt to simulate what the full database in an RPKI world
would look like).  This test repository was a great help when I was
first writing validation code and we were all trying to figure out
what the certificate profile should be, but it was somewhat high
maintenance, and APNIC decommissioned it a while ago.

George Michaelson might be able to supply more details, but as I
recall it was on the order of 30,000 - 40,000 objects.  Didn't include
ROAs or manifests, just certificates and CRLs.  I no longer recall
whether it included EE certificates or only CA certificates.

Most of RPKI objects are quite small.  Disk space per se was never a
problem during testing with the full ~40k objects.  Inodes, on the
other hand, were a problem -- your average unix filesystem is
configured assuming a higher data/inode ratio.  The test objects that
I generate as part of my regression testing framework come in at about
2000 octets for a manifest, about 1250 octets for a certificate, and
about 435 octets for an empty CRL.  These are probably on the small
side, CRLs and manifests grow with the number of entries and
certificates grow with the number of discrete resources encoded as
extensions, but we're still talking about tiny files in most cases.

Manifests issued by RIRs are likely to be significantly larger, simply
because the tree is so flat there.  CRLs sizes depend on your
assumptions, but it's probably safe to assume that anybody who issues
a lot of certificates will need to revoke some of them.

Hope this helps.
_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr

Reply via email to