AI3 Tokyo, JP Meeting AI3 Tokyo, JP Meeting 1 June 2003 June 2003 UNIVERSITI SAINS MALAYSIA IPv6 throughput over IPv6 throughput over Satellite links Satellite links (update) (update) Data collection by: Ezrin, Bukhary, CH Tang Tat Chee Wan ([email protected]) Network Research Group School of Computer Sciences University of Science Malaysia 11800 Minden, Penang, Malaysia
23
Embed
UNIVERSITI SAINS MALAYSIA AI3 Tokyo, JP Meeting 1 June 2003 IPv6 throughput over Satellite links (update) Data collection by: Ezrin, Bukhary, CH Tang Tat.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 11June 2003June 2003
UNIVERSITI SAINS MALAYSIA
IPv6 throughput over IPv6 throughput over Satellite links (update)Satellite links (update)
Data collection by: Ezrin, Bukhary, CH TangTat Chee Wan ([email protected])
Network Research GroupSchool of Computer SciencesUniversity of Science Malaysia
11800 Minden, Penang, Malaysia
June 2003June 2003 AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 22
UNIVERSITI SAINS MALAYSIA
Satellite Link Expt SetupSatellite Link Expt Setup
• USM & Temasek Polytechnic via AI3 USM & Temasek Polytechnic via AI3 networknetwork– 2 hops2 hops– 512 kbps end-to-end (forward and reverse 512 kbps end-to-end (forward and reverse
• Netperf for IPv6 supportNetperf for IPv6 support
• IPv4 and IPv6 TCP/UDP testsIPv4 and IPv6 TCP/UDP tests
June 2003June 2003 AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 33
UNIVERSITI SAINS MALAYSIA
Satellite Link ConfigurationSatellite Link Configuration
512 kbps
512 kbps
512 kbps
1.5 Mbps
1.5 Mbps
512 kbps
1.5 Mbps
USM (4.5m) Temasek Polytechnic (4.5m)
512 kbps
end-to-end
JP (6.0m)
June 2003June 2003 AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 44
UNIVERSITI SAINS MALAYSIA
Terrestrial Link Expt SetupTerrestrial Link Expt Setup
• Terrestrial Link tests done in NRG:Terrestrial Link tests done in NRG:– NRG leased-line to TMNet (512 kbps)NRG leased-line to TMNet (512 kbps)– USM leased-line to Jaring (4 Mbps) USM leased-line to Jaring (4 Mbps)
[Firewall blocks IPv6 traffic][Firewall blocks IPv6 traffic]• IPv4 tests between TMNet and JaringIPv4 tests between TMNet and Jaring• IPv6 tests using tunneling to MANIS IPv6 tests using tunneling to MANIS
(tunnel broker) via TMNet(tunnel broker) via TMNet• FreeBSD 4.7FreeBSD 4.7• Netperf for IPv6 supportNetperf for IPv6 support• TCP tests onlyTCP tests only
June 2003June 2003 AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 55
UNIVERSITI SAINS MALAYSIATerrestrial Link Terrestrial Link
ConfigurationConfiguration
NRG
Jaring ISP
TMNet ISP
MANISIPv6 Tunnel Broker
Node BJaring IP
Node ATMNet IP
IPv6 Tunnelto MANIS
IPv4 Network
4 Mbps
512 kbps
USMNet
June 2003June 2003 AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 66
• FreeBSD 4.5/4.7 has identical performance FreeBSD 4.5/4.7 has identical performance for TCP traffic over IPv4 and IPv6 via for TCP traffic over IPv4 and IPv6 via satellite linksatellite link– TCPv4 for Linux (RH 7.1) appears to scale better TCPv4 for Linux (RH 7.1) appears to scale better
(from previous results)(from previous results)– TCPv6 for Linux (RH 7.1) had worse TCPv6 for Linux (RH 7.1) had worse
performance compared with FreeBSD 4.5: performance compared with FreeBSD 4.5: immature IPv6 stack?immature IPv6 stack?
• UDP performance over satellite is too UDP performance over satellite is too variable to draw any conclusionvariable to draw any conclusion
June 2003June 2003 AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 1414
• Firewall causes TCPv4 to have low throughput Firewall causes TCPv4 to have low throughput – TMNet->Jaring (firewall) low throughput (~150 kbps)TMNet->Jaring (firewall) low throughput (~150 kbps)– Jaring->TMNet high throughput (> ~500 kbps)Jaring->TMNet high throughput (> ~500 kbps)– Averaged value (~350 kbps) for comparison Averaged value (~350 kbps) for comparison
purposespurposes
• TCPv6 performance over terrestrial link still TCPv6 performance over terrestrial link still exceeds that for Satellite linkexceeds that for Satellite link– Terrestrial IPv6 link is tunneledTerrestrial IPv6 link is tunneled– Terrestrial & Sat links are both 512 kbps (end-to-end)Terrestrial & Sat links are both 512 kbps (end-to-end)– ~400 kbps (terrestrial) vs ~150 kbps (satellite)~400 kbps (terrestrial) vs ~150 kbps (satellite)
• UDP performance not tested for terrestrial linkUDP performance not tested for terrestrial link
June 2003June 2003 AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 1515
UNIVERSITI SAINS MALAYSIA
ConclusionsConclusions
• FreeBSD 4.5/4.7 TCP protocol not FreeBSD 4.5/4.7 TCP protocol not affected by v4 vs. v6 transportaffected by v4 vs. v6 transport
• Linux (RH 7.1) seem to have more Linux (RH 7.1) seem to have more aggressive TCPv4 performanceaggressive TCPv4 performance
• TCPv6 performance over high latency TCPv6 performance over high latency links is very poorlinks is very poor
AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 1616June 2003June 2003
UNIVERSITI SAINS MALAYSIA
Thank YouThank You
Q&AQ&A
AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 1717June 2003June 2003
UNIVERSITI SAINS MALAYSIA
Apr 2002 PresentationApr 2002 Presentation
(Previous results)(Previous results)
June 2003June 2003 AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 1818
UNIVERSITI SAINS MALAYSIA
Experimental SetupExperimental Setup
• Linux RH 7.1 for USMLinux RH 7.1 for USM
• FreeBSD 4.3 for Temasek PolytechnicFreeBSD 4.3 for Temasek Polytechnic
June 2003June 2003 AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 1919
UNIVERSITI SAINS MALAYSIA
TCP v4 Stream Test with various Transfer Size
50.00
100.00
150.00
200.00
250.00
300.00
350.00
400.00
450.00
500.00
16 48 80 112 144
Window Size (KByte)
Th
rou
gh
pu
t (K
bp
s)
100KB
500KB
1MB
2MB
2MB (1 Hop)
June 2003June 2003 AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 2020
UNIVERSITI SAINS MALAYSIA
TCP v6 Stream Test with various Transfer Size
50.00
100.00
150.00
200.00
250.00
300.00
350.00
16 48 80 112 144
Window Size (KByte)
Th
rou
gh
pu
t (K
bp
s)
100KB
500KB
1MB
2MB
2MB IPv4
June 2003June 2003 AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 2121
UNIVERSITI SAINS MALAYSIA
UDP Stream Test with 64KB Socket Buffer Size (USM->TP)
50.00
150.00
250.00
350.00
450.00
550.00
0 2000 4000 6000 8000
Message Size (Byte)
Th
rou
gh
pu
t (K
bp
s)
IPv4
IPv6
June 2003June 2003 AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 2222
UNIVERSITI SAINS MALAYSIA
UDP Stream Test with 64KB Socket Buffer Size (TP->USM)
50.00
150.00
250.00
350.00
450.00
550.00
0 2000 4000 6000 8000
Message Size (Byte)
Th
rou
gh
pu
t (K
bp
s)
IPv4
IPv6
June 2003June 2003 AI3 Tokyo, JP MeetingAI3 Tokyo, JP Meeting 2323
UNIVERSITI SAINS MALAYSIA
CommentsComments• TCP Throughput in each direction is consistent TCP Throughput in each direction is consistent
for standard packet size (MTU < 1500 bytes)for standard packet size (MTU < 1500 bytes)• UDP Throughput:UDP Throughput:
– for USM->TP is better for IPv6 (MTU>1500 bytes)for USM->TP is better for IPv6 (MTU>1500 bytes)– for TP->USM is better for IPv4 (MTU>1500 bytes)for TP->USM is better for IPv4 (MTU>1500 bytes)– Different IP Fragmentation setting in routers?Different IP Fragmentation setting in routers?
• IPv6 performance is not optimal over Satellite IPv6 performance is not optimal over Satellite LinksLinks– Need IPv6 stack optimization (KAME?) Need IPv6 stack optimization (KAME?) – TCP throughput is not improved with larger window TCP throughput is not improved with larger window
size size (> 32K): TCPv6 ssthresh not configured?(> 32K): TCPv6 ssthresh not configured?
– Support for jumbo frames?Support for jumbo frames?