Using Traffic Shaping to Combat Spam David Cawley, Senior Engineer December 12th, 2007
Jan 04, 2016
Using Traffic Shaping to Combat Spam
David Cawley, Senior Engineer
December 12th, 2007
Overview
1. Evolution of E-mail & Spam
2. Spamonomics
3. SMTP Multiplexing
4. Traffic Shaping
5. Asynchronous IO
6. Passive OS Fingerprinting
The Dawn of E-mail
• 1965 MIT shared mainframe
• 1971 The @ symbol
• 1976 Queen of England sends an e-mail
• 1982 IETF RFC821/822
• 1989 Lotus Notes released (35k copies sold)
• 1996 Microsoft Internet Mail 1.0
• 2001 IETF RFC2821/2822
Attempts to secure...
• SMTP is inherently insecure
• SMTP-Auth/TLS
• SPF
• Sender-ID
• Why it didn't stop spam
The Evolution of Spam
• 1978 The first spam
• 1988 Usenet cross-posting
• 1993 “spam” coined as a name
• 1997 Open Relays abused
• 2000 Birth of Nigerian spam
• 2001 Formail exploit
• 2003 Sobig virus sends spam
The Evolution of Spam
• 2003 CAN-SPAM act
• 2004 Bill gates prediction & botnets
• 2005 Image spam, Ascii art
• 2006 Animated images, flash, pdf
• 2007 mp3, excel, p2p botnets
The escalating spam problem
Source: spamnation.info/stats
The good old days.
Spammer Economics
• 0.02% people click and buy [source: NY Times]
• Average filter effectiveness is 90%
– 1/10 of spam messages get through
• Improve effectiveness to 95%
– 1/20 of spam messages get through
• Spammer Solution?
– Double spam volume
– Same profit
Traditional Filtering
• MD5's, Fuzzy Signatures, Bayesian
• Header Regex, RBL's, URL Lists, Grey Listing
• Problems
– Obfuscation Techniques
– Formats – html, image, pdf, doc, xls, ole, mp3..
– Zombies, Botnets
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 300
100000
200000
300000
400000
500000
600000
700000
800000
How often do we see a unique Botnet IP?
The Number of Unique IP's versus the number of times reported
# Times Reported
# U
niq
ue
Bo
tne
t IP
's
SMTP Multiplexing
• Transparent SMTP Proxy
• Connection Pooling
• Insulates the MTA
• Avoids delay of legitimate mail
• High Concurrency
– Up to 10,000 simultaneous connections
12
Traffic Shaping
• What can we do?
• Provide a Quality of Service
• Reputation Network
• Throttle unknown senders
• Fast track legitimate senders
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 50 100 150 200 250 300 350 400 450
Perc
enta
ge o
f Con
necti
ons S
till C
onne
cted
Time (Seconds)
Spammers are Less Patient than Legitimate Senders
Spammers
Legitimate Senders
16
Does Sendmail Throttle?
ratecontrol
ConnectionRateThrottle
conncontrol
Asynchronous IO
• Non-Blocking front end
• Blocking Back-end
• Event driven
• Finite State Machine
• Management of Resources
Passive OS Fingerprinting
1.Look at IP packet data
2.Determine the Operating System
3.Decision to Throttle
OS Comparison
Delivered
Windows
Linux
FreeBSD
Solaris
Novell
HP
NetCache
Not delivered
Windows
Linux
FreeBSD
Solaris
Novell
HP
NetCache
Conclusions
1.Spamming is driven by economics
2.Botnet operators need to make money
3.Slowing down spam makes it go away
Nick Shelness, Former CTO, Lotus:“I am able to report that I have been running an instance of
TrafficControl in my own network for four months, and that it has reduced the volume of spam hitting my boundary MTAs on most days
by approximately 95%.”
+1-778-785-6143
www.mailchannels.com