Chair of Network Architectures and Services Department of Informatics Technical University of Munich
Key Performance Indicators of TCP Flows Patryk Brzoza Advisor(s): - - PowerPoint PPT Presentation
Key Performance Indicators of TCP Flows Patryk Brzoza Advisor(s): - - PowerPoint PPT Presentation
Chair of Network Architectures and Services Department of Informatics Technical University of Munich Key Performance Indicators of TCP Flows Patryk Brzoza Advisor(s): M.Sc. Simon Bauer Supervisor: Prof. Dr.-Ing. Georg Carle Technical
2
Patryk Brzoza | Key Performance Indicators of TCP Flows
Agenda
Introduction Protocol Background KPI Overview Measurement Methods Data Processing Deriving Network Events Conclusion
3
Patryk Brzoza | Key Performance Indicators of TCP Flows
Introduction
- Studies: 95% of total traffic volume provided by TCP [1]
- TCP Flows play crucial role for overall performance
How to evaluate this and which metrics are important?
4
Patryk Brzoza | Key Performance Indicators of TCP Flows
Protocol Background: Transport Control Protocol (TCP)
- Reliable and connection-oriented communication
- Connection: two unidirectional flows
5
Patryk Brzoza | Key Performance Indicators of TCP Flows
KPI Overview
- Performance of TCP constrained by its features
Deriving metrics: Key Performance Indicators (KPIs)
Latency Indicators TCP Flow KPIs Throughput Indicators Other Indicators Packet Loss Indicators
6
Patryk Brzoza | Key Performance Indicators of TCP Flows
KPI Overview: Latency Indicators
Temporal quality and responsiveness metrics of flows Round Trip Time (RTT):
- Interval between sending segment and receiving
acknowledgement Connection Setup Time (CST):
- Interval for establishing
connection
7
Patryk Brzoza | Key Performance Indicators of TCP Flows
KPI Overview: Packet Loss Indicators
8
Patryk Brzoza | Key Performance Indicators of TCP Flows
KPI Overview: Throughput Indicators
9
Patryk Brzoza | Key Performance Indicators of TCP Flows
KPI Overview: Other Indicators
Metrics that do not fit in previous categories Response Time (RT):
- Processing time between last packet of response and first packet
- f request
Reset Rate (RT):
- Relation of RST segments to overall transmitted segments
10
Patryk Brzoza | Key Performance Indicators of TCP Flows
Measurement Methods
- Data must be measured first before processing it:
Measurement Methods Active Passive Generate probe segments Capture segments from link ✘ ✔
11
Patryk Brzoza | Key Performance Indicators of TCP Flows
Measurement Methods
- How many monitors should be set up and at which location? [3]
Algorithms can determine good solutions
12
Patryk Brzoza | Key Performance Indicators of TCP Flows
Measurement Methods
- Possible measurement setup [4]:
- Differentiation between bidirectional and unidirectional
measurement
13
Patryk Brzoza | Key Performance Indicators of TCP Flows
Measurement Methods
Bidirectional approach:
- Extract KPIs by correlating segments
Unidirectional approach:
- Determine flow type first [4]:
Feedback Download Unknown
?
14
Patryk Brzoza | Key Performance Indicators of TCP Flows
Data Processing
- Measurements can generate vast amounts of data
- Use data mining approaches to reduce it [5]:
− Find relationships − Filter out unnecessary packets
15
Patryk Brzoza | Key Performance Indicators of TCP Flows
Data Processing: RTT calculation
16
Patryk Brzoza | Key Performance Indicators of TCP Flows
Data Processing: RTT calculation
Unidirectional measurements: − SYN-based method: determine segment distance depending on flow type − Flight method: group packets by inter-arrival time, measure distance of groups
17
Patryk Brzoza | Key Performance Indicators of TCP Flows
Network Event Derivation
- KPIs give valuable information for intrusion detection, e.g.:
− Throughput spikes: flood-based DoS attacks [5]
− High Reset Rate: port scan attacks
- Can also expose errors and misconfigurations
18
Patryk Brzoza | Key Performance Indicators of TCP Flows
Conclusion
- TCP flow performance constrained
- Measurable with various classifiable KPIs
- Collect data with either active or passive measurements
- Reduce data amount with mining approaches
- Extract KPIs from bi- or unidirectional measurements
- Detect intrusion attacks or network errors using KPIs
19
Patryk Brzoza | Key Performance Indicators of TCP Flows
Questions? Thank you for your attention! Any questions?
Contact: brzoza@in.tum.de
20
Patryk Brzoza | Key Performance Indicators of TCP Flows