Developing Web100 Based Network Configuration

1
Developing Web100 Based Network Configuration
Performance Measurement Tools

• Web-100 Workshop
• Aug 1, 2002
• by
• Rich Carlson
• Argonne National Laboratory

2
Motivation for work

• Develop single shot diagnostic tool that
  doesnt us historical data
• Measure performance to users desktop
• Combine numerous Web100 variables to analyze
  connection
• Develop network signatures for typical network
  problems

3
ANLs Web100 Based Approach

• Single test to gather end-to-end data
• Gather data from multiple variables
• Compare measured performance to analytical values
• Translate network values into plain text messages

4
Different HW same Network port

• 10 Mbps NIC
• Jan 25 093045 146.137.160.253 6826 6781 Kbs
  5840 16382 B 0 20 ms 25 3 2
• 100 Mbps NIC
• Jan 25 100557 146.137.160.252 84557 86514 Kbs
  5792 63712 B 0 10 ms 0 0 0

5
Sample results

• Speed
• 94.09
• 22.50
• 82.66
• 33.61
• 6.99
• 7.15

• 100 Mbps FD
• Ave Rtt loss
• 5.41 0.00
• 1.38 0.78
• 6.16 0.00
• 14.82 0.00
• 10 Mbps
• 72.80 0.01
• 8.84 0.75

6
Sample results

• Speed
• 94.09 Good
• 22.50 Bad NIC
• 82.66 Bad reverse
• 33.61 Congestion
• 6.99 Good
• 7.15 Bad NIC

• 100 Mbps FD
• Ave Rtt loss loss/sec
• 5.41 0.00 0.03
• 1.38 0.78 15.11
• 6.16 0.00 0.03
• 14.82 0.00 0.10
• 10 Mbps
• 72.80 0.01 0.03
• 8.84 0.75 4.65

7
ANL developed tools

• Web based tool
• based on ORNL tool
• modified analysis messages
• Command line based tool
• modified version of NLANRs Iperf ver 1.2
• supports 2 new options
• -e (-ee) print Web100 derived details
• -R reverse direction of data flow
• modified analysis messages

8
ANL developed tools

• Geared toward campus area network
• End-user based view of network
• Provides some hard evidence to users and
  network administrators to reduce finger pointing
• Can be used to identify performance bottlenecks

9
Web Based Performance tool

• Operates on Any client with a Java enabled Web
  browser
• What it can do
• Positively state if Sender, Receiver, or Network
  is operating properly
• Provide accurate application tuning info
• Suggest changes to improve performance

10
Web base Performance tool

• What it cant do
• Tell you how a 3rd party is performing
• Tell you where in the network the problem is
• Tell you how other clients will perform

11
Network Signatures

• Duplex Mismatch Detection
• Connection is Congestion limited
• CWNDtime gt 30
• Have experienced packet loss
• MaxSsthesh gt 0
• Retransmitting packets
• PktsRetrans/sec gt 2
• Throughput is less than expected
• estimate gt 2

12
Network Signatures

• Full/Half Duplex Detection
• Connection is Receiver buffer limited
• RWINtime gt 95
• Connection transitions between Receiver limited
  and Sender limited
• RwinTrans/sec gt 30
• SendTrans/sec gt 30

13
Current Deployment

• 3 servers at ANL
• Miranda Externally visible
• Ophelia, Cordelia ANL Internal only
• Servers support both Web based tool and Iperf
  based tool
• Tools available via anonymous ftp from
  achilles.ctd.anl.gov/pub/web100-tools

14
Usage Statistics (July 22, 2002)
15
Future plans

• Better estimate of path Bandwidth
• receiver side packet pair arrivals
• ACK packet timings
• Explore receiver side statistics
• Explore Gigabit Ethernet issues
• Identify partners/collaborators

16
Caveats

• Server needs time between tests (90 sec)
• Analysis messages need to be validated

17
Demo

• http//miranda.ctd.anl.gov7123

18
Obtaining the test results

• Runs 10 sec test from Client to Server
• no diagnostic data collected
• Runs 10 sec test from Server to Client
• Web100 diagnostic data collected at end of test
• Prints out summary status message
• Link speed and duplex
• Informational or Warning messages

19
Analyzing the test results

• Statistics button
• Send and Receive throughput achieved
• Details for 5 configuration tests (link type,
  duplex mode, congestion, excessive errors, duplex
  mismatch condition)
• Throughput limits section (S-R-N limited, RTT,
  loss, out-of-order)
• Tweakable settings (TCP modifications to
  improve performance)

20
Analyzing the test results

• More Details button
• Individual TCP counters collected by Web100
• Conditional test parameters
• Throughput analysis section including theoretical
  limits, bandwidthdelay products, loss rate, and
  buffer sizes

21
Analyzing the test results

• Report Problem button
• Invokes local email client ltmailtogt
• Automatically inserts collected data into body of
  email
• Provides comment section for user feedback
• Server logs all counter variables used for
  condition tests