Re: Problems with implementation - DoS attacks possible

At 12:59 PM 1/22/2003, Mikael Olsson wrote:


>Naohiro Fukuda wrote:
> >
> > Biggest problem is to collect all those messages. But, if we
> > implement the router preparing several key paris and
> > signing/storing them to reduce the time for signing, then the
> > performance problem of the router will get better, then creating
> > traceback message as to include hash, key, and signature, the
> > message will be handled at collector without waiting for the
> > another key discorsure message.
>
>Um. I still don't think you've understood the problem.
>
>Big backbone routers that forward several gigabit of data each second
>CANNOT afford to pubkey sign each packet itrace packet.
>
>Your suggestion of "preparing" keys doesn't help.  Even if we prepare
>1 million keys, on such backbone routers, this fails after only a
>few seconds when the key list has been exhausted; we're back to
>generating keys on-the-fly, which, again, they don't have the CPU for.

Or to put it a better way: It could certainly be made to work, but would 
require an "itrace coprocessor" that takes care of all the tasks related to 
this matter. While not out of the question, it would require additional 
hardware.


>The situation is just as bad on small edge routers. They don't come
>equipped with pentium class CPUs. They come with small CPUs that are
>dimensioned for management logic only. They do packet processing in
>custom ASICs. So, even if the data stream is much smaller, they still
>don't have the oomph to sign everything.

Actually in most cases this isn't true even today. Many of the edge routers 
don't contain ASICs and do all their processing in software on a general 
purpose CPU. Depends what you consider a "small edge router" I like the 
26xx series as a typical edge/CPE box, which contains a PowerPC core in 
which ALL of the forwarding and management work occurs.



>And, even if through some Great Miracle, all routers suddenly started
>emitting pubkey signed itraces, what kind of hardware do we want to
>require at the collectors?  We want people with "normal" internet
>connections to be able to do this on the equivalent of a standard PC.
>We do NOT want to require boxes equipped with crypto accelerators.

Here I'll disagree with you a bit too. The typical P4-powered PC on a 
desktop sits idle nearly all the time. It's quite capable of crunching the 
keys.

I won't disagree with your basic premise that the issue of signing the 
packets is a problem. I do, however, think it should be possible to 
convince both router vendors and their customers of the need for extra 
components in routers to keep the Internet viable, should they prove necessary.

I am concerned that if the cryptography is too compute intensive, the 
itrace mechanism may wind up providing its own DoS against its users. That 
would be bad. However, if there is a need down the road to provide a high 
level of cryptography to prevent attacks against itrace, then vendors 
should start thinking about how to get there.

We should explore whether it is possible to deploy an itrace mechanism with 
SOME level of security and with low compute overhead in the near term and 
continue to work on ways to strengthen the cryptography further.