ietf-asrg
[Top] [All Lists]

Re: [Asrg] 7. Best Practices - DNSBLs - Article

2003-08-13 13:32:37
Yakov Shafranovich wrote:

Chris, can you provide some information on DNSBLs and how your system evaluates their effectiveness?

Taking the second part first:

1) Our false positive process involves having the sender contact a specific address, and that will contain a "block ID" from the message rejection. Using the block identifier, we can retrieve the message, and identify _why_ it was blocked (may not be DNSBL). Then, if it's a DNSBL, we can locally whitelist the IP.

2) Our metrics scan the logs and identify block/unblock counts for all messages by IP. During metrics processing, we generate tables of how much email a given blacklist blocked, _plus_ how many email messages _would_ have been blocked by that blacklist, but weren't because they were whitelisted.

Thus, for each blacklist, we generate:

a) how many emails were blocked, and the percentage of total inbound it represents. b) how many emails had to be whitelisted for that blacklist, and what percentage of the emails blocked by that blacklist it represents.

Note that the latter (b) isn't _really_ a "lost email" metric (to us). Those messages weren't lost, they got through because we whitelist. But it's a good FP metric.

Basically coming up with a "blocking percentage" and a "whitelisted" percentage for each blacklist.

Here's our overall effectiveness metrics - last two week period - percentages are of _total_ email to our inbound mail servers.

This is meant merely as a demonstration of what significant site _should_ be doing with metrics from their own filtering, not an exhaustive list of all DNSBLs, or what others see.

Due to the variety of DNSBLs, it's very difficult to generate overall metrics for all. Without zone transfering the DNSBL for example, it's too slow to measure it with reasonable sample sizes.

The counts are _all_ total recipients affected. So, a single email with 5 recipients counts as 5.

Rule                        count     %       FPs  % of blocked
CBL                        352153  16.86       7   0.0020
MONKEYPROXY                261436  12.52      83   0.0317
CONTENT                    260391  12.47     N/A
BOPM                       167277   8.01       0   0
OSsocks                    152168   7.29       0   0
SPEWS (not used)            94285   4.51     N/A
NTthresh                    54233   2.60      27   0.0498
SBL                         47614   2.28     156   0.3276
NTmanual                    45366   2.17    2955   6.5137
NTauto                      25094   1.20      12   0.0478
OBproxies                   21068   1.01      18   0.0854
PDL                         16547   0.79     N/A
NTnordns                     5814   0.28       0
Flonetwork                    229   0.01       0

[The below numbers are whitelisted recipients per associated BL, as a percentage of the total inbound, not of blocks per-se. "OK" is _total_ whitelisting entries, whether or not the whitelisting is still necessary]:

OK                          36770   1.76
OK SBL                        156   0.01
OK OBproxies                   18   0.00
OK NTthresh                    27   0.00
OK NTmanual                  2955   0.14
OK CBL                          7   0.00
OK NTauto                      12   0.00
OK MONKEYPROXY                 83   0.00

TOTAL                     2088400 100.00
TOTAL BLOCK                571850  27.38

Ie: out of 2088400 attempted deliveries, 352153 (16.86%) were blocked by the CBL. Only 7 messages (0.0020% of 352153) would have been blocked if we weren't whitelisting.

Some notes:

"CONTENT" is the union of all of our non-DNSBL based. Our system doesn't let us compute FPs on that directly.

BOPM, Monkeys, OSSocks, OBproxies are third party open proxy/socks blacklists. The first three are public.

CBL is more-or-less an open proxy/socks blacklist (see cbl.abuseat.org)

SBL is Spamhaus.

NTmanual is local manual black/whitelist (the high FP rate is perfectly okay because those are whitelisted, and it's under our control)

NTnordns is "spam over threshold, no rDNS" (AOL does this now)

NTthresh is "spam over threshold, listed in ORDB or SpamCop"
NTauto is local open relay/proxy detector.

Our system doesn't let us compute FPs on BLs we don't use, so we don't have the number for SPEWs.

PDL (DHCP pools) does block occasional legitimate email, but, by policy we don't want email from DHCP pools. So we don't whitelist it - we get the sender to either use their proper mail servers, or get PDL to update the entry if it's erroneous. Whitelisting would be silly if the entry was accurate, because the IP can essentially be random.

Flonetwork: zone listing doubleclick/dartmail/flonetwork. No, we _really_ don't want email from them, period.

If someone wants subjective commentary on the public DNSBLs (ie: proposed BCP compliance), let me know.

Some information in this area would be useful as well.

there's been a few of those posted.


_______________________________________________
Asrg mailing list
Asrg(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/asrg