Re: [Dots] (draft-ietf-dots-signal-filter-control) ACL Stats issue

Tiru, 

I agree that we are approaching the problem with two different use cases:

(1) a client domain with is basically "consuming" services. I do still think this use case does not need to learn about the ACL stats. 

(2) your case in which the client domain is "providing" services: I still think that the impact on business can be determined also using local information (known patterns + rate-limit policy applied by the client). If the goal is to decide whether/when an alternate mitigator is to be solicited, this can deterministically rely upon "status" set to 4 (Attack has exceeded the mitigation provider capacity) or deactivate back the rate-limit ACL + local observation. Please remember that local observation is needed for efficacy update.

Cheers,
Med

> -----Message d'origine-----
> De : Konda, Tirumaleswar Reddy [mailto:TirumaleswarReddy_Konda@McAfee.com]
> Envoyé : lundi 29 avril 2019 12:18
> À : Jon Shallow; BOUCADAIR Mohamed TGI/OLN; kaname nishizuka; dots@ietf.org
> Objet : RE: [Dots] (draft-ietf-dots-signal-filter-control) ACL Stats issue
> 
> > -----Original Message-----
> > From: Jon Shallow <supjps-ietf@jpshallow.com>
> > Sent: Monday, April 29, 2019 3:11 PM
> > To: Konda, Tirumaleswar Reddy
> > <TirumaleswarReddy_Konda@McAfee.com>;
> > mohamed.boucadair@orange.com; kaname nishizuka <kaname@nttv6.jp>;
> > dots@ietf.org
> > Subject: RE: [Dots] (draft-ietf-dots-signal-filter-control) ACL Stats issue
> >
> > This email originated from outside of the organization. Do not click links
> or
> > open attachments unless you recognize the sender and know the content is
> > safe.
> >
> > Hi,
> >
> > See inline,
> >
> > Regards
> >
> > Jon
> >
> > > -----Original Message-----
> > > From: Dots [mailto: dots-bounces@ietf.org] On Behalf Of Konda,
> > > Tirumaleswar Reddy
> > > Sent: 29 April 2019 10:22
> > > To: mohamed.boucadair@orange.com; Jon Shallow; 'kaname nishizuka';
> > > dots@ietf.org
> > > Subject: Re: [Dots] (draft-ietf-dots-signal-filter-control) ACL Stats
> > > issue
> > >
> > > > -----Original Message-----
> > > > From: mohamed.boucadair@orange.com
> > > <mohamed.boucadair@orange.com>
> > > > Sent: Monday, April 29, 2019 2:46 PM
> > > > To: Konda, Tirumaleswar Reddy
> > > <TirumaleswarReddy_Konda@McAfee.com>;
> > > > Jon Shallow <supjps-ietf@jpshallow.com>; 'kaname nishizuka'
> > > > <kaname@nttv6.jp>; dots@ietf.org
> > > > Subject: RE: [Dots] (draft-ietf-dots-signal-filter-control) ACL
> > > > Stats issue
> > > >
> > > > This email originated from outside of the organization. Do not click
> > > > links or
> > > open
> > > > attachments unless you recognize the sender and know the content is
> > safe.
> > > >
> > > > (Focusing on this particular point).
> > > >
> > > >
> > > > > -----Message d'origine-----
> > > > > De : Konda, Tirumaleswar Reddy
> > > > > [mailto:TirumaleswarReddy_Konda@McAfee.com]
> > > > > Envoyé : lundi 29 avril 2019 10:52 À : BOUCADAIR Mohamed TGI/OLN;
> > > > > Jon Shallow; 'kaname nishizuka'; dots@ietf.org Objet : RE: [Dots]
> > > > > (draft-ietf-dots-signal-filter-control) ACL Stats issue
> > > > >
> > > > > > > >
> > > > > > > >  ACL-specific stats and mitigation stats will give a clear
> > > > > > > > > picture of the traffic rate-limited, bad traffic dropped
> > > > > > > > > by the DDoS mitigation system, and using these stats the
> > > > > > > > > DOTS client can heuristically determine the amount of
> > > > > > > > > legitimate traffic dropped because of rate-limit and the
> > > > > > > > > impact of the attack
> > > on its
> > > > service.
> > > > > > > >
> > > > > > > > [Med] The impact can be observed locally (e.g., bad QoS,
> > > > > > > > inability to
> > > > > > > access a
> > > > > > > > service, instable connectivity, etc.). I still don’t see how
> > > > > > > > sharing the
> > > > > > > ACL stats
> > > > > > > > will be helpful here.
> > > > > > > >
> > > > > > > > A DOTS client can preinstall the same rate-limit filter with
> > > > > > > > but with
> > > > > > > different
> > > > > > > > policies. It can select the appropriate ACL to
> > > > > > > > activate/deactivate based on local experience.
> > > > > > >
> > > > > > > I don't get how the local experience will help the client pick
> > > > > > > an alternate mitigation provider who can handle the attack scale.
> > > > > >
> > > > > > [Med] Modern CPEs include automated features to assess the
> > > > > > availability of services such as VoIP, IPTV, etc. The DOTS
> > > > > > client can be fed with input
> > > > > from
> > > > > > these modules and react accordingly.
> > > >
> > > > [Med] s/client/server.
> >
> > Is this correct?
> >
> > > >
> > > > >
> > > > > I meant the target network cannot infer the amount of legitimate
> > > > > traffic (or infer the number of users) unable to use its service
> > > > > because of the rate- limit action.
> > > > >
> > > >
> > > > [Med] The amount of traffic is not required to assess the
> > > > availability of "nominal" services (the example above). What is
> > > > really important is
> > > whether
> > > > some critical services are available. That information can be
> > > > determined
> > > without
> > > > needing the ACL stats.
> > >
> > > I am not referring to "nominal" services or critical resources. For
> > > instance, consider Netflix is not accessible to a large number of
> > > users because of the rate-limit action.
> >
> > The DOTS server will have a limited (only because they have to be
> previously
> > defined) set of (possibly inactivated) ACLS on the server.  If the
> "standard"
> > white/black list are unable to bring the inbound pipe back to not being
> > flooded, then a (likely global for the DOTS client's networks) Rate-Limit
> ACL
> > must be brought in.  Once the Inbound pipe is available, then analysis of
> the
> > data reaching the DOTs client will show the top users which then need their
> > own limiting (black or rate-limit) ACL set up over the data channel.  At
> this
> > point the Rate-Limit ACL can removed to see if things are stable again.
> > [I agree that the CPE may not have this top usage capability]
> >
> > If Netflix (or similar) has a priority when under attack, then this needs
> to be
> > added into a White ACL which can be done once the inbound pipe is not
> > flooded (or be a part of the standard white lists)
> 
> I think we are discussing two different use cases. My attack use case is
> Netflix content provider is under volumetric DDoS attack, and if the rate-
> limit ACL is configured using the DOTS signal channel because the DDoS
> mitigation provider cannot handle all the attack traffic. The rate-limit ACL
> stats will help Netflix understand the scale of the attack, impact on the
> current business because of the rate-limit action (e.g. based on the amount
> of traffic dropped by DMS, infer the amount of good traffic dropped by the
> rate-limit ACL action, and infer the number of users who cannot access its
> service), and if the attack lasts for  several days/weeks help identify an
> alternate mitigation provider capable of handling the attack (e.g. Krebs was
> initially using Akamai and eventually got protected by Google to handle the
> massive attack).
> 
> Cheers,
> -Tiru
> 
> >
> > ~jon
> >
> > >
> > > -Tiru
> > >
> > > _______________________________________________
> > > Dots mailing list
> > > Dots@ietf.org
> > > https://www.ietf.org/mailman/listinfo/dots