On 11/Jul/11 22:11, John Levine wrote:
>>> On diagnosing a failure, the agent generates a random number R in the
>>> interval [0, 1] (or sets R=0.5). It then computes a value P using numeric
>>> values from the retrieved RR and a formula to be specified in
>>> authfailure-report. If P >= R, then the agent generates and sends the
>>> report, otherwise does nothing.
>
> Standards only work when it is mutually beneficial to both parties to
> implement them. What is the benefit to the reporting party of
> implementing this complex mess?
Complex mess?!? It can be done by a 40-line C function that I attach
untested. The benefit is the simplicity of being stateless while
still not requiring to report each and every failure.
The alternative needs a semaphore to synchronize access to global
store. Then, what kind of global store? Plain files like
/tmp/reports-example.com? MySQL? BerkeleyDB? Note that a similarly
structured storage may already exist, e.g. maintained by the local
identity assessor, but sharing it with the verifier would add even
more complexity.
> If there's any rate limiting at all, keep it to something easy to
> implement, like a maximum number of reports per hour.
Yes, such limit makes much sense, either as a global maximum or as a
per-domain one. Shouldn't it be set by the reporter, though? In such
case it could be enforced /after/ any self-imposed "complex mess"
limit, e.g. on sending, where access to locally maintained databases
might be easier.
#include <stdlib.h>
#include <math.h>
int self_limit(char *ri, unsigned long ttl)
/*
assume ri points to the value of the ri= parameter, if any, and
that the parameter should be formatted like "86400p2s100d" in any
order (p, s, d stand for for Period, Slope or Sharpness, Divisor)
*/
{
unsigned long p = 0, s = 0, d = 0;
double f = 1.0;
while (ri && *ri)
{
char *t;
unsigned long l = strtoul(ri, &t, 10);
switch (*t)
{
case 'd': d = l; break;
case 'p': p = l; break;
case 's': s = l; break;
default:
return 0; // invalid ri, don't send
}
ri = t + 1;
}
if (s && p && ttl) // Gaussian cutoff
{
double x = (double)s * (1.0 - (double)ttl/(double)p);
if (fabs(x) >= 1.0)
return 0; // don't send
f = M_E * exp(-1.0/(1.0 - x*x));
}
if (d) // linear cutoff
f /= (double)d;
return f * (double)RAND_MAX <= rand();
}
_______________________________________________
marf mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/marf
