Case for TaskDriven NetworkWide Abstraction approach to Enterprise Design

1
Case for Task-Driven Network-Wide
Abstraction approach to Enterprise Design

• Yu-Wei (Eric) Sung
• Internet Systems Lab
• Dept of ECE, Purdue University
• (with Prashant Garimella, Sunil Krothapalli, and
  Sanjay Rao)

LISA 2007 Configuration WorkshopNov 12, 2007,
Dallas, TX
2
Motivation Task-driven Network-wide Abstractions

• Tremendous interest in simplifying enterprise
  design and configuration using abstractions
• Many prior efforts
• Template-driven (eg Presto), BGP Policy
  specification (eg RSPL), Vendor-neutral config
  specification (eg SNMP MIBS)
• Our proposal Task-driven Network-wide
  abstraction approach to network design
• Task-driven capture intended requirements, eg
  performance, security, manageability
• Network-wide consider requirement of the network
  as a whole
• Focus on enterprise campus networks ? not well
  understood

3
Methodology

• Three-pronged research framework
• First step White-Box approach to study
  enterprise networks
• Static analysis of topology and router/switch
  configurations
• Gain deeper understandings through interaction
  with operators
• White-box studies are rare
• Significant effort to bootstrap relationships
  with operators
• Sensitive nature of data access non-trivial

Bottom-Up, White-Box study
Top-DownNetwork Design
Abstraction
4
Case Study VLAN

• Why study VLAN?
• Widely prevalent in enterprise/campus networks,
  little understanding in research community
• VLAN Configuration time-consuming, error-prone
• Critical area for abstraction
• Prevalent _at_ Purdue University
• 200 routers, 1300 switches, hundreds of VLANs
• First empirical study of VLAN usage (SIGCOMM INM
  Workshop 2007)

5
What is VLAN?

• A VLAN groups hosts attached to different
  switches as if they are on the same wire
• Each VLAN belongs to a stand-alone subnet
• Simplifies address assignment, e.g. writing ACLs
• VLAN spans at Purdue University
• 80 1-2 buildings some entire campus (e.g.
  classrooms)

.
.
.
.
R1
S
R2
Trunk
S3
S2
S1
VLAN 3
.
Access
.
.
.
H2
H3
H4
H1
VLAN 2
VLAN 1
VLAN 2
VLAN 1
6
3 key components of VLAN Configuration
interface Vlan1 description subnet
192.168.1.0/24 ip address 192.168.1.1
255.255.255.0
interface FastEthernet0/1 switchport mode trunk
switchport trunk allowed vlan 1,2
R1
S
interface FastEthernet0/1 switchport mode
access switchport access vlan 1
Fa0/2
Fa0/1
Fa0/2
S1
Fa0/1

• Access port
• Trunk port
• Subnet Config

H1
H2
VLAN 1
VLAN 2
7
Design Issues

• Communication between hosts from different VLANs
  must go through a router
• Performance Inefficiencies Longer delay, harder
  to debug since data may flow through other
  buildings
• Placement of designated router Where to best
  place the router for hosts in a given VLAN?

? Building1
Data
R2
R1
S
Trunk
S3
S2
VLAN 3
S1
.
Access
.
.
.
Physical
H2
H3
H4
H1
VLAN 2
VLAN 2
VLAN 1
VLAN 1
8
Results

• Performance Inefficiencies
• Inefficiency hops in data flow path/ hops
  in physical path
• Inefficiencies as large as 14, gt 4 for about 12
  host pairs
• Placement of designated router
• Sub-optimal placement a router not placed in the
  same building as where the majority of hosts in
  that VLAN are
• More pronounced for wider-spread VLANs 11, 33,
  and 58 for VLANs that span 1, 2, and gt2
  buildings
• Typical cause evolution of network
• See INM07 paper for complete results and
  discussions of other issues

9
Example Task-driven VLAN Abstraction and
Top-down Network Design

• Abstraction Complete network topology consists
  of routers/switches/hosts, a set of hosts in VLAN
  X
• Suggest a router for VLAN X
• Task 1 Minimize inefficiencies communicating
  with all other VLANs
• Task 2 Task 1 Host H in VLAN X is a major
  network storage server for the entire network

10
Summary Future Work

• Goal Task-driven, Network-wide abstraction to
  enterprise network design
• Three-pronged research framework
• First step White-box approach to studying
  networks, providing critical insights for
  abstraction design
• VLAN as a case study
• Issues inefficiencies, sub-optimal router
  placement
• First study of VLAN usage in a real network
  (INM07)
• Going Forward
• Formulate VLAN abstractions and demonstrate their
  applications
• Generalizing other campus/enterprise networks

11
Thank you

• We are looking for data!
• Please let me know if you have router/switch
  configurations of an entire enterprise/campus. We
  will provide useful tools in return. ?
• Email sungy_at_purdue.edu
• Visit http//www.ece.purdue.edu/isl for more
  info about our research
• Questions?