On Tue, Jan 23, 2007 at 11:32:28AM -0500, Larry J. Blunk wrote:
> Erik Romijn wrote:
> >You have a point there. If one takes dump files from random sources, it
> >is not possible to determine where they came from.
> >
> >On the other hand, it would add bloat...
> >
>     If the table dumps were done in a single atomic MRT record
> (something I've thought about), I think this would be reasonable.
> [...]
>    I don't think people are generally in the habit of pulling down
> table dumps from random unknown sources.    And in the end,
> the IP/AS of the collector is not terribly critical information.

I agree. Let's not bloat it more.

>   That said, we could perhaps define an initial header MRT
> record for table dumps that includes the collector's IP and
> AS number

That may be useful. But still, there is your good point that people will
not fetch random dumps and then try to find out where they came from.

> perhaps an indexed array of peer IP/AS numbers. 

Nice idea. But how much would we save?

Let's assume asn16/ipv4, so 48 bits of IP/AS per entry. With a bit more
than 200K routes that's 10Mbit of IP/AS.
With the index, assuming a 16-bit number and 200 peers, it would be 9600
bits of array and 3.2M of indexes. So that's 33% cut (on peer
information only).
We save about 0.85Mbyte here. A RIS RIB dump on the biggest machine is
240M uncompressed. So we would save 0.35%.

Assuming best-case, ipv6/asn32. Without array 30Mbit, so that's 0.7% save
based on RIS examples.

With an 8-bit index it's 0.7% for asn16/ipv4 and
1.4% for asn32/ipv6. However, the latter allows only 255 peers.

So the reduction in data is really small. Unless I miscalculated
ofcourse :-)

> >>> The View and Status fields have been dropped
> >>>      
> >
> >What about a router with multiple bgp instances?
> >Wasn't the views field intended to be able to make the distinction
> >between entries from the different bgp instances?
> >
> >(this is also solved by adding Peer IP and AS)

>       Don't understand the comment about Peer IP and AS as those
> are already in the record.

I mixed up. I was referring to the collector's own IP and AS, which
doesn't make any sense either. Forget that comment :-)

> >>>[..] peer AS size (16 or 32 bit) [..]
> >>>      
> >>why bother with 2 byte ASNs?
> >>    
> >
> >Indeed. Why not make Peer AS always 4 bytes?
> >  
> 
>  Again, bloat.   There's plenty of bits available in the subtype
> code field, why not use them?   Lot's less CPU cycles/time
> to do a simple bit test for field size than to transfer around
> and decompress unneeded bits.

I must have missed somewhere that asn16/asn32 would be encoded in the
subtype field.
And yes, if we store the bits (subtype) anyway, your idea is a lot more
efficient as we won't have much asn32* for the coming time.

> >And if we stick to this, how would a parser know the boundary between
> >peer AS and Prefix Length?
> 
>   Bits are defined in the subtype code to define both peer
> IP address size and AS number size.   You know where the
> Prefix Length field begins based on the size of the peer's
> IP address and AS number.

Ofcourse. If it's encoded in the subtype, the answer to my question is
very obvious :-)


* Probably the correct term here should be "4-byte only AS", as AS2134
  can be used on an asn32 BGP implementation too, but would be stored as
  16-bit number.

cheers,
-- 
Erik Romijn                 RIPE NCC jr. software engineer
http://www.ripe.net/        Information Services dept.

Attachment: pgp3qUh6acBKm.pgp
Description: PGP signature

_______________________________________________
GROW mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/grow

Reply via email to