de title
m 48 pt
ubtitle
m 30 pt
A compressive method for
maintaining forwarding states
in sdn controller
Ying Zhang*, Sriram Natarajan$, Xin Huang^, Neda
Beheshti*, Ravi Manghirmalani*
Ericsson Research*, NTT$, CYAN^
de title
m 32 pt
2 rows
level 1
m 24 pt
vel 2-5
m 20 pt
YZ[\]^_`abcd
£¤¥¦§¨©ª«¬®¯°
ÕÖ×ØÙÚÛÜÝ
÷øùúûüýþÿĀ
ģĪīĮįİıĶķĹĺĻļĽ
ŠšŢţŤťŪūŮů
ẃẄẅỲỳ–
ĞĠĠĢĢĪĪĮĮİĶĶĹ
ŢŢŤŤŪŪŮŮŰŰ
ΤΥΦΧΨΪΫΆΈΉΊ
ΎΏ
ЛМНОПРСТУ
ЛМНОПРСТУ
ЎЏѢѢѲѲѴѴҐ
ects or
er area
Controller needs to maintain a copy of
forwarding states
›  Forwarding states in SDN: tables on the switches
›  Most existing controllers do not maintain a copy of
forwarding states
›  There is a need in real-world deployment scenario
– Fast fault recovery from transient switch failures
– Network state queries by multiple control applications
– Consistency check on the switches’ actual states
– Rule space analysis for optimization, debugging, etc.
de title
m 32 pt
2 rows
level 1
m 24 pt
vel 2-5
m 20 pt
YZ[\]^_`abcd
£¤¥¦§¨©ª«¬®¯°
ÕÖ×ØÙÚÛÜÝ
÷øùúûüýþÿĀ
ģĪīĮįİıĶķĹĺĻļĽ
ŠšŢţŤťŪūŮů
ẃẄẅỲỳ–
Need an efficient solution to store states in
controller
›  Naïvely storing all the tables from all switches takes huge
space
›  Explore correlation between switches
– Intuition: because of topological properties, there exists dependency
between actions of different switches
a
ĞĠĠĢĢĪĪĮĮİĶĶĹ
ŢŢŤŤŪŪŮŮŰŰ
ΤΥΦΧΨΪΫΆΈΉΊ
ΎΏ
ЛМНОПРСТУ
ЛМНОПРСТУ
ЎЏѢѢѲѲѴѴҐ
ects or
er area
b
dst = a, sent to S2
dst = b, sent to S2
dst = c, sent to S2
dst = a, sent to S3
dst = b, sent to S3
dst = c, sent to S3
c
de title
m 32 pt
2 rows
level 1
m 24 pt
vel 2-5
m 20 pt
YZ[\]^_`abcd
£¤¥¦§¨©ª«¬®¯°
ÕÖ×ØÙÚÛÜÝ
÷øùúûüýþÿĀ
ģĪīĮįİıĶķĹĺĻļĽ
ŠšŢţŤťŪūŮů
ẃẄẅỲỳ–
ĞĠĠĢĢĪĪĮĮİĶĶĹ
ŢŢŤŤŪŪŮŮŰŰ
ΤΥΦΧΨΪΫΆΈΉΊ
ΎΏ
ЛМНОПРСТУ
ЛМНОПРСТУ
ЎЏѢѢѲѲѴѴҐ
ects or
er area
Use prediction tree to capture dependency
Match
fields
S1 fwd
to port
S2 fwd
to port
S3 fwd
to port
1.2.1.0
1
1
1
2.1.1.0
2
1
2
1.2.2.0
1
1
1
2.1.2.0
2
1
2
1.2.3.0
3
2
1
2.1.3.0
3
2
2
1.2.4.0
4
2
3
2.1.4.0
4
2
3
S1’s fwd port number <=2
y
n
S2’s port=1
S2’s port=2
S1’s fwd port number <3
n
y
Match field<2.0.0.0
y
S3’s port=1
n
S3’s port=3
S3’s port=2
de title
m 32 pt
2 rows
level 1
m 24 pt
vel 2-5
m 20 pt
YZ[\]^_`abcd
£¤¥¦§¨©ª«¬®¯°
ÕÖ×ØÙÚÛÜÝ
÷øùúûüýþÿĀ
ģĪīĮįİıĶķĹĺĻļĽ
ŠšŢţŤťŪūŮů
ẃẄẅỲỳ–
ĞĠĠĢĢĪĪĮĮİĶĶĹ
ŢŢŤŤŪŪŮŮŰŰ
ΤΥΦΧΨΪΫΆΈΉΊ
ΎΏ
ЛМНОПРСТУ
ЛМНОПРСТУ
ЎЏѢѢѲѲѴѴҐ
ects or
er area
System design and results
Table
combination
-  Handling empty
entries
-  Handling multiple
tables
Dependency
discovery
-  Use Bayesian
network to identify
dependencies
between columns
-  Build classification
and regression tree
(CaRT)
Compression
-  Divide columns to
predicting and
predicted sets
-  Support
incremental update
›  Evaluation on Mininet with NOX
›  Using Bayesian network and C4.5 decision tree for dependency
discovery
›  Results:
–  Regular topology: 1.77% of original size
–  Realistic topology: 2.8% of original size
de title
m 32 pt
2 rows
level 1
m 24 pt
vel 2-5
m 20 pt
Q&A?
YZ[\]^_`abcd
£¤¥¦§¨©ª«¬®¯°
ÕÖ×ØÙÚÛÜÝ
÷øùúûüýþÿĀ
ģĪīĮįİıĶķĹĺĻļĽ
ŠšŢţŤťŪūŮů
ẃẄẅỲỳ–
ĞĠĠĢĢĪĪĮĮİĶĶĹ
ŢŢŤŤŪŪŮŮŰŰ
ΤΥΦΧΨΪΫΆΈΉΊ
ΎΏ
ЛМНОПРСТУ
ЛМНОПРСТУ
ЎЏѢѢѲѲѴѴҐ
ects or
er area
6