Re: Destination-IP-Origin-AS Filter for BGP Flow Specification

Jeffrey Haas <[email protected]> Fri, 8 Nov 2019 12:11:06 -0500
Newsgroups gmane.ietf.idr
Message-ID <[email protected]>
Haibo,

Thanks for the expected clarification.

To the chairs, as noted below the feature violates some fundamentals of flowspec architecture.  I'd strongly object to the feature being adopted without a new SAFI to get it out of the existing flowspec use cases.

-- Jeff


> On Nov 7, 2019, at 8:48 PM, Wanghaibo (Rainsword) <[email protected]> wrote:
> 
> Hi Jeffrey,
> 
> 	Flowspec is designed for security protection, but current usage is not limited to security protection, 
>    but also to optimize traffic. Using Flowspec to optimize traffic is safer and more reliable than send routes to router.
> 
>    Our scenario here is for optimize traffic. The actions here are all performed on the routers. You need to perform a FIB lookup to get dest-as.
> 
>    This does violate some of the current rules and introduces a certain performance penalty,
>    but it can reduce the demand for Flowspec entry space, so you can use this rule in the appropriate scenario.
> 
> Regards, 
> Haibo
> 
> -----Original Message-----
> From: Jeffrey Haas [mailto:[email protected]] 
> Sent: Friday, November 8, 2019 1:38 AM
> To: Wanghaibo (Rainsword) <[email protected]>
> Cc: Robert Raszuk <[email protected]>; [email protected]; Zhuangshunwan <[email protected]>; idr@ietf. org <[email protected]>
> Subject: Re: [Idr] Destination-IP-Origin-AS Filter for BGP Flow Specification
> 
> Haibo,
> 
> On Tue, Nov 05, 2019 at 08:11:12AM +0000, Wanghaibo (Rainsword) wrote:
>> PS: Netflow is already supporting statistical traffic based on Dest-IP-Origin-AS, it already download Dest-IP-Origin-AS to FIB entry, this prosess can be reused.
> 
>> From a forwarding perspective, this is the detail that bothers me.
> 
> Flowspec right now is currently independent of FIB state.  It functions on the firewall layer, which is typically implemented prior to FIB.
> 
> What this feature implies is something roughly like:
> Packet comes in, hits rule to check dst-as.
> dst-as lookup needs to happen as one of:
> - communicate to BGP routing process. (not likely to scale)
> - trigger a FIB lookup, check returned dst-as.  Violates pipelining in many
>  architectures.
> - Have the entire FIB's dst-as-map pushed into memory for firewall,
>  implement as a longest-match lookup on that collection.
> 
> -- Jeff

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