High Speed Optical Interconnect Project May08-06

1
High Speed Optical Interconnect ProjectMay08-06
2
Team Information
Client Lockheed Martin

• Team Members
• Team Leader Adam Jackson
• Communication Coordinator
• Nick Ryan
• Bader Al-Sabah
• David Feely
• Richard Jones

• Faculty Advisor
• Dr. Ahmed Kamal
• Client Contacts
• Aaron Cordes
• Rick Stevens

3
Problem Statement

• At this time, the maximum real-world throughput
  of 10 Gbps network configurations is unknown.

Approach

• Investigate a 10 Gbps network using fiber optical
  components
• Test final system

4
Requirements

• Non-Functional
• Must be implemented before the class completes in
  May 2008.
• Must determine financial feasibility of fiber
  optic components.

• Functional
• The network shall be implemented with 10 Gbps
  components.
• The team shall test the network to determine real
  throughput.

5
Functional Continued

• 2.1 The team shall test the network to determine
  real-world bandwidth.
• 2.2 The team shall test the network to determine
  bandwidth efficiency.
• 2.3 The team shall test the network to determine
  switch time.

• 2.4 The team shall test the network to determine
  latency.
• 2.5 The team shall test the network to determine
  quality of service.

6
Project Risks

• A team member could leave the project
• Insufficient funding for project completion
• Component failure

7
Deliverables

• The team shall present the chosen design solution
  to LM for approval before implementation.
• The team shall provide weekly reports to LM.
• The team shall present the test results to LM.
• The team shall deliver a final report and all
  documentation to LM.
• The team shall deliver the prototype to LM at the
  completion of this project.
• All deliverables shall be presented to Rick
  Stevens on or before May 1, 2008.

8
Design Strategy
Design Strategy Block Diagram
9
Market Research

• µTCA

• PCI-E

• Advantages
• Modular design allows for expansion
• 262.5 Gbps maximum throughput for Advanced
  Mezzanine Cards (AMC)
• Disadvantages
• AMC Network Interface Cards at 10 Gbps are not
  readily available
• Costly components

• Advantages
• Readily available optical 10 Gbps NICs
• Variety of 10 Gbps XFP Switches
• Relatively low cost components
• Disadvantages
• Lack of PCI-E systems at ISU

Source http//www.compactpci-systems.com/columns/
Tutorial/pdfs/4.2005.pdf
10
System Concept
11
Switched Configuration

• Composed of three nodes and a Ethernet switch
• Used for testing switching time, latency, and
  quality of service

Graphic inspired by previous HSOI team
12
Hardware Setup
Myricom 10G-PCIE-8A-RE 10 Gigabit Ethernet
PCIe Adapter with pluggable XFP optical
interface (http//www.myri.com/Myri-10G/10gbe_solu
tions.html)
XFP Transceiver
PCI-E x8 or x16 bus
TigerSwitch 10G 8-Port Standalone XFP 10Gigabit
Ethernet Managed Layer 2 Switch SMC Networks,
Inc.
13
Testing and Analysis Software

• ping
• Used as a latency measurement tool
• iperf
• Bandwidth measurement tool
• tcpdump
• Packet capture program
• Scripts
• Scripts will control and execute programs
  according to test specifications
• Other
• C program used to analyze results

14
Test Specifications

• Bandwidth
• Compare link usage for each node under varying
  workload types
• Bandwidth Efficiency
• Show a comparison of the amount of OSI Layer 1
  data sent for different OSI Layer 7 data block
  sizes
• Latency
• Compare the latency between nodes under different
  network loads

15
Test Specifications

• Switching Time
• Compare switching time and link load for cases
  when 2 and 3 nodes are connected to the network
• Quality of Service
• Show the amount data received from each sending
  node for each endpoint node over time

16
Bandwidth Test Results

• 30 iperf runs
• Command Line iperf -p 5001 -c 192.168.1.2
• Max 1.39
• Min 1.38
• Avg 1.38968
• Standard Deviation 0.001796
• 99 Confidence Interval 1.38893,1.39043

17
Bandwidth Efficiency Calculations
18
Latency Test Results

• 1000 ping runs
• ping -c 5 192.168.1.2
• Max 0.035 ms
• Min 0.046 ms
• Avg 0.038145
• Standard Deviation 0.046895
• 99 Confidence Interval 0.034697, 0.041592

19
Resource Costs
Resource Quantity Estimated Cost Actual Cost
Optical NICs 3 3000 2146.50
XFP Switch 1 6500 N/A
XFP Transceiver 6 3720 2700
Fiber optic cables 3 240 130.94
Host System 3 500 Department Systems1
Total 13960 4987.442
1 ISU ECpE Departments update of the Senior
Design lab may cover this cost 210 Shipping
charge for NICs and Transceivers added to total
20
Planned Work Breakdown
Personnel Meetings Research Design Implementation Testing and Debugging Documentation Website Totals
Adam 80 20 8 20 32 10 1 171
Bader 80 20 8 20 32 10 1 171
Nick 80 20 8 20 32 10 1 171
David 80 20 8 20 32 10 1 171
Richard 80 20 8 15 30 10 15 178
Total 400 100 40 95 158 50 19 862
21
Actual Work Breakdown
Personnel Meetings Research Design Implementation Testing and Debugging Documentation Website Totals
Adam 73.75 34 2 7 14.75 21.25 0 152.75
Bader 81 29 0 1 5.5 5.5 0 122
Nick 79.5 25.5 4 1 3.75 19.5 0 133.25
David 60.5 30 0 8.5 5 11 0 115
Richard 79.25 44 1.5 0 6 12 18.75 161.5
Total 374.05 162.5 7.5 17.5 35 69.25 18.75 684.5
Summary of Individual Hours
22
Project Completion Analysis
Estimated Percent Completion Budgeted Hours Budgeted Hours Completed
Meetings 100 400 400
Research 95 100 95
Design 100 40 40
Implementation 95 95 80.75
Testing 75 158 118.5
Documentation 95 50 47.5
Website 97 19 18.5
Table used to calculate Budgeted Cost of Work
Performed
23
Project Analysis

• Budgeted Cost of Work Scheduled
• 862 scheduled hours 10 per hour 8,620
• Actual Cost of Work Performed
• 684.55 actual hours 10 per hour 6,845.50
• Budgeted Cost of Work Performed
• 800.25 scheduled hours 10 per hour 8,002.50

24
Earned Value Analysis

• Cost Variance
• 8,002.5 - 6,735 1,267.5
• Under budget
• Cost Performance Index
• 8,002.5 / 6,735 1.188
• Under budget

• Schedule Variance
• 8,002.5 - 8,620 -617.5
• Behind schedule
• Schedule Performance Index
• 8,002.5 / 8,620 0.928
• Behind schedule

25
Accomplishments

• Researched 10 Gbps network architectures
• Completed test scripts
• Completed analysis scripts and programs
• Executed copper tests and analyzed results
• Executed limited fiber optic tests

26
Realized Risks

• Unable to obtain fiber optic switch
• PCI-E slot not fully compliant
• First system ran at x1
• Second system was graphics only

27
Lessons Learned

• How to coordinate buying components with various
  groups.
• Prepare for unexpected sources of error.
• Fully research the technologies before we use
  them.

28
Conclusions

• Creating an affordable fiber optic network is
  feasible.
• Copper results illustrate proof of concept on
  testing approach.
• Preliminary results using PCI-E x1 speeds show
  that bandwidth efficiency is poor unless
  parameters are changed.

29
Questions?

• ?