Trace collection in the UNCCH DiRT lab

1
Trace collection in the UNC-CH DiRT lab
The UNIVERSITY of NORTH CAROLINA at CHAPEL HILL
2
Outline

• Monitoring location and UNC network
• Setup specs - infrastructure
• Process sequence overview
• Live demo
• Post-processing
• Our experimental infrastructure and its
  capabilities

3
Trace collection point
UNC-CH network
Internet Internet2
Monitor w/ DAG card
4
Infrastructure

• Monitor specs P4 1.8 GHz CPU, 1.4 GB RAM, 532GB
  hard drives, FreeBSD 4.7.
• DAG 4.3GE dual interface 1000Base-SX Gigabit
  Ethernet 133MHz PCI-X card.

5
Process sequence overview
convert to tcpdump format using dagconvert
packet capture using dagsnap
anonymize using tcpdpriv
create connection vectors in tmix
run (100s of) tmix experiments in the lab
analyze and present results
6
Demo

• /usr/local/bin/dagsnap -d /dev/dag0 -s 300 -v -o
  test.dag
• dagconvert -Terfpcap -i ../test.dag -f b gt
  test.b.tcpdump
• dagconvert -Terfpcap -i ../test.dag -f a gt
  test.a.tcpdump
• /usr/dirt/src/tcpdpriv-1.1.10/tcpdpriv -P99 -A60
  -M20 -t mapping_file -r test.a.tcpdump -w
  test.a.anon.tcpdump
• /usr/dirt/bin/4.x/tcpdump-hiperf -I -S -n -tt -r
  test.a.anon.tcpdump tcp gt test.a.anon.tcpdump.txt
  (-I ? Print IP header fields (ttl, id and
  length))

7
Packet capture

• Usage dagsnap options
• -d ltdevicegt DAG device name
• -h display help (this page)
• -j Maximize disk write
  performance - will only write in chunks
• -m ltmebibytesgt Maximum amount of data to
  write per call in MiB (default 4)
• -o ltfilenamegt output file name (default is
  stdout)
• -s ltsecondsgt runtime in seconds
• -v increase verbosity
• With -v three columns are printed per second.
• 1. The cumulative total of data written out.
• 2. The buffer occupancy. Small values
  indicate no packet loss.
• 3. The rate at which data is currently being
  written.

8
Convert dag to tcpdump

• dagconvert DAG file conversion utility Usage
  dagconvert options
• -d ltdevicegt DAG device name
• -h display help (this
  page)
• -v increase verbosity
• -i ltfilenamegt input file
• -o ltfilenamegt output file
• -r Nkmg change output file
  after N Bytes.
• k, m, g suffixes for
  kilobytes, megabytes, gigabytes.
• -s ltsnaplengt output snap length
• -t ltsecondsgt capture period in
  seconds
• -T ltin_typeout_typegt input and output types
  (see list of types below)
• -f ltlistgt comma separated list
  of filters (see list of filters below)

9
anonymize

• usage /usr/dirt/src/tcpdpriv-1.1.10/tcpdpriv
• P99 pass TCP port numbers through unchanged
• A60 map address using mapping file
• M20 option for multicast addresses

10
Example plot throughput (Mbps)

• A 10-min trace collected at low-usage time
  (230am)

11
Example plot throughput (Kpps)
12
Analysis

• Simple statistics like throughput plots over time
• Loss rates
• RTT (round trip time) calculations
• Running tmix experiments with the data
• analyze results
• create various possible scenarios and test (e.g.
  what does doubling the traffic throughput do to
  end user response times?)

13
Network Topology
Monitors
Ethernet Switch
Ethernet Switch
Router
Router
End systems
End systems
14
Tmix discussion

• Replay trace on the network in a controlled
  environment
• Throughput can be varied up or down
• Induce losses and delays per flow
• Replaying a-b-t sequences