Transcript PPT - Microsoft Research

Self-Managed Networks:
Dream or Reality?
Jawad Khaki
Corporate Vice President
Windows Networking & Device Technologies
Current Situation
Management is expensive
Devices only understand low-level
settings
Diagnostics/monitoring is primitive
Need a comprehensive network
solution
ISP, hotspot
Enterprise
Home
IT Budgets
Pain Points
Complexity due to inconsistency
Heterogeneous world
Different configuration models
Variety of monitoring techniques
Version/vendor specific repair procedures
Hard to understand dependencies
Networking problems are a significant cause
of overall service failure (Oppenheimer,
USITS’03)
Network causes 15% of all problems resulting in
downtime (Forrester survey of IT pros)
Not humanly solvable
Operator error is largest cause of service
failures in some environments
(Oppenheimer, USITS’03)
40% of downtime is due to human
operators (Candea, ’03)
In many environments, operator may not
be tech savvy (e.g., home) or even
immediately available (e.g., space, sensor
nets).
Consumer networking support calls are
time consuming, e.g., power cycle
router/modem = avg 53 min (MS PSS)
End-to-End Approach Essential
Apps/users understand behavior
desired
Network admins understand high-level
design goals/constraints
The dream is to integrate end-user
knowledge and administrative goals
Big Dreams
Self-managing networks
Self-deploying and self-cleaning
Self-configuring and self-adapting
Self-optimizing
Self-protecting
Self-monitoring
Self-diagnosing
Self-healing
Prevention more than cure
A self-* system requires knowledge of itself and its
environment, it is self-aware
Some Real Examples Today
Policy distribution systems allow autodeployment of configuration across a
network
Routing protocols auto-adapt to
topology changes and failures
TCP auto-adapts to congestion
Demos
Product Engineering
Challenge
Design for experience
End user: Focus on the task not technology
Network manager: Design, deploy, operate
Must get the fundamentals right
Essential to think through scenarios
Work flow
Intelligence
Environment
Always keeping the customer in mind
Hard issues
Multiple administrative
organizations
Different relationships
Peers
Customer-provider
Arbitrary
Lack of trust motivates privacy
constraints
Unaligned goals means configuration
is a challenge
Possibility of catastrophic
failure
Defect in automation can have disastrous
results
“Rogue equipment can create a monster
headache. It can easily waste a million dollars of
resources.”
-IT admin, large LA corporation
Broadcast storms due to protocol or software
bugs (Spurgeon, 1989)
One router vendor tried offering automated
config repair features, but found that customers
were afraid to deploy it
Possibility of exploitation by malware
Tension between control and
automation
Flexibility of business models and
preferred treatments
Compliance requirements
Job security for operators
Natural aversion to loss of control
Change to unfamiliar technology
Need to find the right
balance
Policy to express high-level
constraints
Self-management within those
constraints
Static routes
Static addresses
etc
Control
BALANCE
Dynamic routing
Dynamic addresses
etc
Automation
Summary
Innovation in fundamentals just as important
as new scenarios
Make secure, effortless, reliable, efficient
operation the forethought
Let humans succeed at what they’re good at
Let’s solve the hard issues
Dealing with heterogeneity of
device types and vendors
Hard to visualize existing state and
dependencies
Expensive to maintain multiple
configuration/monitoring systems
Need for common solutions
Simplicity
Heterogeneity
Dealing with poorly written
applications
“Some applications need to know what machine
a person is on...we found that giving the
docking stations a static IP address and the
laptop a static IP address makes it easier for
us.” (IT Admin, Medium Org, New York)