Hi Bill,
Yes, we agree that if apps always used names, the problem
would at least be confined to the network stack. I thought
that was already fairly clear, but we'll work your points in.
Thanks for the SMTP point, it needs to be added.
Brian
On 2008-10-28 03:03, William Herrin wrote:
> On Thu, Oct 23, 2008 at 4:41 PM, Brian E Carpenter
> <[EMAIL PROTECTED]> wrote:
>> Ran and I drafted this in reaction to the earlier discussion here
>> about the impracticality of renumbering. We'd like comments and, above
>> all, contributions. For now, this list is suggested for any discussion,
>> but I will also mention the draft on the intarea list.
>>
>> http://www.ietf.org/internet-drafts/draft-carpenter-renum-needs-work-00.txt
>
>
> Brian,
>
> You missed the more fundamental DNS problem:
>
> 95% of the software which interacts with the DNS does so via the
> gethostbyname call. Gethostbyname does not return the TTL for the
> information, thus the applications which use it do not conceive of a
> hostname whose address changes during the course of the application's
> operation.
>
> Even if you were to deprecate and eventually eliminate the
> gethostbyname call, painless renumbering is prevented by the rest of
> the API which expects the layer-7 application to interact with the
> layer-3 address. Because the application works right 99.9% of the time
> (when IP addresses aren't changing) few small-time applications have
> or will implement the TTL logic.
>
> Some very large systems get it wrong too. One obvious example: Windows
> XP and Windows Vista cache DNS query results for 15 minutes regardless
> of the TTL reported by the DNS resolver. Windows 98 cached DNS query
> results for 24 hours or until the next reboot.
> (http://support.microsoft.com/default.aspx?scid=KB;en-us;263558).
> Before you think to blame all our woes on Microsoft: In Solaris 8 and
> earlier, nscd also ignores the DNS TTL, passing already-stale records
> back to apps which call gethostbyname.
>
> Anecdotal reports from web server administrators suggest that a
> hostname-based requests for web sites continue to trickle in at the
> old IP address for months after the server address has changed in the
> DNS.
>
> Bottom line: because the APIs call for the applications to interact
> with the layer-3 addresses directly instead of interacting with only
> hostname targets, painless renumbering is an unattainable goal.
>
>
> You also need a heading: SMTP issues.
>
> Nearly all of the trust-management systems (aka anti-spam systems) for
> email derive from the server IP address. It tends to go haywire when
> the address of any large originator of email changes, already a source
> of much pain and consternation for system administrators.
>
> Regards,
> Bill Herrin
>
>
_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg
