North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: DoS attacks, NSPs unresponsiveness

  • From: dies
  • Date: Thu Nov 02 22:16:46 2000

	I agree that getting help from Tier1/2 NSPs is not a outlandish
request/idea.  However, in my personal experience UUNet has been the most
responsive out of all the Tier-1 NSPs (Assuming they are contacted
correctly).  Tracking the attack, however, is not always a trivial
task.  If you are a large NSP like UUNet, Sprint, and C&W, where your
backbone spans multiple POPs or even countries with their own POPs it is
time consuming and you run a real risk of crashing transit routers if the
flood is heavy (read high packets per second).  At that point the question
becomes, "Is it worth tracing an attack on a shell server/irc
server?"  I'm not trying to say running an IRC server or shell server is
wrong, however, some (if not most) NSPs/ISPs consider them attractive

	This topic is continously covered (as you can tell) so I'm not
sure where to go from here.  I've been looking into starting a BGP speaker
that announces the top 4000 smurf amplifiers from my list and the top 2000
from netscan, which in turn can be pushed to null0 or discard.  This would
prevent one from actually attacking from a network participating in this
BGP session.  I've tested it and it's working on a couple smaller ISPs as
we speak.  My major problem is getting some Tier-1's to go for
it...Anyways I guess larger attacks than last Feb. will have to happen to
get something done?  Let's hope not...

On Fri, 3 Nov 2000, Ariel Biener wrote:

> On Thu, 2 Nov 2000, dies wrote:
>   Actually, I was thinking of taking a slightly different path.
> 1). ISPs (downstreams) responsibilities:
> 1a). Active and clueful NOC.
> 1b). Proper implementation of Internet RFCs (private networks and
>      spoofing).
> 1c). Proper BGP filtering towards upstream.
> 1d). Running routing software adequate to today's challenges (for example,
>      something that can deal properly with small fragments...).
> 2). NSPs (upstreams, tier1/2) responsibilities:
> 2a). Active and clueful abuse department.
> 2b). Monitoring tools that will allow tracking down a stream throghout
>      their own backbone, and cooperation with their peers to get to the
>      source. With a handful of cluefull people, and a set of autmatic
>      tools, this IMHO is something at hand.
> 2c). Proper BGP filtering of their customers, to not allow their customers
>      to fuck-up Internet routing (Sprintlink incidents...).
> The main idea is that downstreams make sure they are not very susceptable
> to abuse. Then, if attacked, the upstream provider should cooperate in
> detecting the source, by monitoring their own network, and cooperating
> with it's peers. If for any reason, it takes more than X hours, the
> upstream provider should try to protect it's customers. Usually, it wont,
> and the attacker will be indentified, and shut down at it's source.
> There is no requirement that tier1/2 NSPs tie up their equipment into an
> ACL monster. What is needed is cooperation. With proper cooperation,
> identifying a stream's source is much easier that you may think.
> Do you find the above so hard to do ?  All it requires is some
> professional attitude, and a bit more cooperation and consideration from
> NSPs and ISPs as one, what happens to someone you don't know today may
> happen to you tomorrow !
> Thoughts ?
> --Ariel
> > 	Well since everyone else is stating their opinions, I'll join in
> > as well.First off I think pulling the plug is a great idea ( =] ).  
> > Anyways the point comes down to this.Who should be doing the ingress
> > filtering?Tier-2's, Tier-1's, the actual customer?I know this whole
> > idea sounds very pretty and nice, however, when it comes down to it there
> > are many real problems with this idea.One, the hardware on most ISP's
> > backbones cannot realistically do ingress filtering.I'm sorry to say but
> > a GSR is not able to do ingress filtering on 5 Channelized OC-12's that
> > hold 400+ Customers a piece.It just does not work, I don't care what
> > Cisco claims, it just does not work.What about other vendors?  I have no
> > experience with Bay or Lucent, however, Juniper (which I do have
> > experience with) has the ability due to the hardware based filtering
> > available but that brings up a whole set of other questions.How will
> > ingress filtering from an ISP level effect downstream customers that do
> > asymmetrical routing?How about the management overhead that comes into
> > play when you are a Tier-1 or a large Tier-2 with tens of thousands of
> > customers?What is comes down to is that customers need to be doing
> > egress filtering, it's the only scalable solution, however this just is
> > not happening.Don't blame the ISPs only, it's their customers that are
> > really the problem.Lack of security/knowledge on the customer's end
> > leads to hacked boxes, which in turn lead to DoS attacks.It really comes
> > down to not the responsibility of the ISP, but in fact the responsibility
> > of the customers!Maybe we all should thinkg about that before we point
> > fingers.
> > 
> > 
> > 
> > On Thu, 2 Nov 2000 [email protected] wrote:
> > 
> > > On Thu, 02 Nov 2000 09:59:04 EST, MarkMentovai <[email protected]>said:
> > > > This can't go on forever.I'd like to spread the clue about ingress
> > > > filtering, and am willing to commit time to the cause.Is anyone with me?
> > >
> > > The problem is that for many ISPs, I fear the only way to get them to
> > > implement 2827-style filtering is for their upstreams to implement a
> > > policy of fascist-mode ingress filtering - "We see a bogon packet that
> > > your site should have filtered, we pull the plug on your link till you
> > > fix it".
> > >
> > > Time alone won't be enough.Bring a baseball bat.  And a spare bat.
> > >
> > > --
> > > 				Valdis Kletnieks
> > > 				Operating Systems Analyst
> > > 				Virginia Tech
> > >
> > >
> > >
> > 
> > 
> --
> Ariel Biener
> e-mail: [email protected]           Work phone: 03-6406086
> fingerprint = 07 D1 E5 3E EF 6D E5 82 0B E9 21 D4 3C 7D 8B BC