Why dont most users see high performance

1
Why dont most users seehigh performance?
Setting the Stage Panel NASA/NGI Gigabit
Networking Workshop 15 Aug 2000

• Phillip Dykstra
• WareOnEarth Communications Inc.
• phil_at_wareonearth.com

2
TCP Throughput Once window size and available
bandwidth arent the limit

• 0.7 Max Segment Size (MSS)
• Bandwidth lt
• Round Trip Time (latency) sqrtloss
• M. Mathis, et.al.
• Double the MTU, double the throughput
• Halve the latency, double the throughput
• Halve the loss rate, 40 higher throughput

3
Delay Matters - Light in Fiber
10 ms
20 ms
30 ms
40 ms
Actual rtts often 1.4 - 3.5x longer
4
Receive Windows at 1 Gbps
1 MB
8KB limit is 4 miles
2 MB
3 MB
4 MB
5 MB
5
Loss Limits for 1 Gbps
7x10-7
2x10-7
7x10-8
4x10-8
MSS 1460
6
Maximum Transmission Unit (MTU) Issues
kc, Miller, Thompson, 1998

• Dont accept the new 1500 byte standard
• Insist on GigE Jumbo Frames

7
There are a lot of bugs

• 100 Mbps is hard to get
• Recent treno test show poor results
• Mperf shows a lot of anomalies
• Ethernet duplex problem epidemic

8
Some MPing Results
9
Summary - To Reach 1 Gbps

• Tune end systems (make rwin adaptive)
• Minimize delay (approach line of site)
• Reduce loss (nearly loss free)
• Increase Packet Sizes (9KB)
• Fix countless bugs