Re: Has PSI been assigned network 1?
From: bmanning@ISI.EDU
avg> RIPE database does not control the actual routing.
Nor does any other registry today.
Sorry, this is factually wrong. Merit PRDB does control ANS routing. That is the only reason why it is used at all. avg> I'm not sure the current RADB project has any mention avg> of real-time updates of gated from the database. avg> With dozens of updates we do every day it nearly as good avg> as useless.
The last customer request was for 5 minute intervals, which is well within the existing bounds for BGP convergance time, at least as reported by you for at least one Sprint box.
Did i mention that resetting BGP every 5 minutes is a sure way to KILL internet? It got to be incremental updates, not the "generate gated config and then reload it entirely".
Following Dales comment, each provider may set its own parameters on generation of configurations, regardless of where it gets its data from.
Sorry, the only available technology today is heavily taxed even w/o doing massive filtering. --vadim
Vadim: I'm not speaking for Bill Manning (I could never manage that at all ;-)).. but 90% of configuration changes in the PRDB are network based. Less are AS pathway based. Each of these could generate a new configuration file, and have it reloaded, but **not** drop the bgp sessions. As you said -- incremental updates for bgp, but the config files are different. Were you worried about the bgp restart or the implementation of the configuration file? Both have thoughts and solutions... from the RA team - MERIT and ISI. Your earlier comments on the requirements for the RADB have been most helpful. Any further insight would be very helpful. It's nice to know what user's desire of the RS. Boy... I'd like to have Sprint as a satisfied customer. Sue Hares
In message <199504201748.NAA23695@titan.sprintlink.net>, Vadim Antonov writes:
From: bmanning@ISI.EDU
avg> RIPE database does not control the actual routing.
Nor does any other registry today.
Sorry, this is factually wrong. Merit PRDB does control ANS routing. That is the only reason why it is used at all.
avg> I'm not sure the current RADB project has any mention avg> of real-time updates of gated from the database. avg> With dozens of updates we do every day it nearly as good avg> as useless.
The last customer request was for 5 minute intervals, which is well within the existing bounds for BGP convergance time, at least as reported by you for at least one Sprint box.
Did i mention that resetting BGP every 5 minutes is a sure way to KILL internet?
It got to be incremental updates, not the "generate gated config and then reload it entirely".
Gated processes the configs without dropping the BGP session. At least it is designed to do so and with enough physical memory to keep the box from paging it will. Gated is capable of digesting a new 200,000 lines configuration file and determining what it needs to do about the differences in a minute or less. It then takes a few minutes to go about carrying out the changes and does so without dropping peerings, but it really has to strain to do it if the OS has to move pages in and out of memory to do it.
Following Dales comment, each provider may set its own parameters on generation of configurations, regardless of where it gets its data from.
Sorry, the only available technology today is heavily taxed even w/o doing massive filtering.
Agreed.
--vadim
Curtis
participants (3)
-
Curtis Villamizar -
Susan Hares -
Vadim Antonov