Hello Vince,

That makes a sense.

Is this publicly documented somewhere?


Damien Saucez

Damien Saucez

On 13 Aug 2012, at 22:18, Vince Fuller <[email protected]> wrote:

>> As you certainly know, the mapping system has been migrated
>> to LISP-DDT with success on 3/14/2012. As the migration was
>> planed, we have decided to measure it! To do so, we used
>> vantage points around the world in different network types (EID
>> space, commercial Internet, research Internet...). These vantage
>> points have measured for a period of about one month the
>> mapping system. To do so, they have send a Map-Request
>> for all the EID prefixes with lig, and this every 15 minutes.
>> We expected to observe a big difference of delay and mappings
>> between ALT and DDT. However, the result we obtain is that
>> the change is not very significant. There is no particular loss
>> of map-Request/Map-Reply during the transition and the delay
>> is not significantly increased or reduced. However, we observe
>> a much more variable delay now with DDT than before where 
>> delays were very stable with time.
> 
> Hi Damien-
> 
> Thanks for the interesting analysis.
> 
> I'll offer a simple explanation for why lookup delay on the pilot network
> has become more variable with the deployment of DDT: we have a somewhat
> pathalogical configuration on the pilot network. In particular, the pilot
> DDT topology includes regional DDT Map Servers in that can accept 
> registrations from every ETR in a region. The ETRs in the region, though,
> are only configured to register to a subset of the DDT Map Servers in that
> region. When a DDT Map Request is sent to a DDT Map Server that doesn't
> have a particular ETR registered, a Map-Referral message with action code
> "MS-NOT-REGISTERED" is returned, which causes the requestor to re-try to
> a different DDT Map Server in the current referral set and results in
> at least one additional DDT message round-trip-time.
> 
> This "misconfiguration" is deliberate and is in place for internal testing
> and configuration management purposes. Such a configuration would not be
> used on a "real world" network so variable lookup performance on the pilot
> network should not be considered representative of how things would work
> in operational deployment.
> 
> Hope this helps.
> 
>       --Vince

_______________________________________________
lisp mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/lisp

Reply via email to